Skip to content

Latest commit

 

History

History
1638 lines (1498 loc) · 88 KB

changes.md

File metadata and controls

1638 lines (1498 loc) · 88 KB

weewx-loopdata change history

8.0.7 2026/11/07

  • Archive records seeded into hour and the rolling windows at startup carry their real weight.
  • LoopData runs on a station whose [StdArchive] leaves out archive_interval, which WeeWX allows.
  • Internal: the period and rolling-window accumulator builders no longer take archive_interval, whose only use was that weight.

8.0.6 2026/11/06

  • A failure while handling a loop packet costs that packet, not the file. Any exception there -- a write of loop-data.txt failing because /dev/shm was full, say -- used to end LoopData's processing thread, so loop-data.txt stopped updating while weewxd carried on, until weewxd was restarted. Now that packet and its upload are dropped and the next packet is handled as usual. The first failure is logged with its traceback ("Could not process a loop packet"), and the first success after a run of failures is logged too ("LoopData is processing loop packets again").
  • A blank timeout in [[RsyncSpec]] (timeout = with no value, or none) means the default of 1 second on WeeWX 5.0 and later. It stopped weewxd at startup, even with enable = false. (Before 5.0, WeeWX itself rejects a blank number, here as everywhere else.)
  • LoopData warns about WeeWX's own [[RSYNC]] option names in [[RsyncSpec]]. A port, server, user or path there was silently ignored, since LoopData's names are remote_port, remote_server, remote_user and remote_dir; each now gets a warning at startup naming the option to use instead.
  • A fresh install's [[RsyncSpec]] shows remote_port and ssh_options as commented-out examples.

8.0.5 2026/11/06

  • remote_port in [[RsyncSpec]] works. LoopData handed the port to WeeWX's rsync code as a string, which WeeWX cannot format, so the first upload failed with "TypeError: %d format: a real number is required, not str" and LoopData stopped writing loop-data.txt until weewxd was restarted. Every station on WeeWX 4.6 or later that set remote_port was affected (4.6 changed how WeeWX writes the port into the ssh command); one that left it unset, or put the port in ssh_options or ssh_config, was not. A port set here overrides a Port that ssh_config sets for the host, 22 included.
  • Internal: weewxd's shutdown is recognized by the running weewxd's own Terminate class rather than by the class name, so a library exception that happens to be called Terminate is no longer mistaken for it.

8.0.4 2026/11/04

  • Times declared in unix_epoch_ns are correct on every WeeWX. LoopData took the factors that turn an epoch into milliseconds or nanoseconds from WeeWX, whose nanosecond factor was a million (the microsecond factor) from 2021 until WeeWX 5.6 corrected it; LoopData now owns the arithmetic in both directions, making the value with a thousand and a billion to the second and dividing by the same, so a nanosecond field is a nanosecond count and renders as the right instant on WeeWX 4.6 through 5.6 alike. A page that divided the old value by a million to get seconds was correcting for the bug and now divides by a billion; formatted times are unchanged. Before 5.6 WeeWX's own report tags still publish a microsecond count under that name, so a page should read LoopData's. The same factors cover a unit segment (almanac.sunrise.unix_epoch_ns), the instant at which a next_* almanac group expires, and a time WeeWX's almanac gives as a Dublin julian day (its moon, planet and next_* events, on every WeeWX), which reaches the epoch unit by WeeWX's day-to-seconds conversion and LoopData's factor from there.
  • The sample report declares its clock as current.dateTime.unix_epoch.raw, pinned to seconds, where it declared current.dateTime.raw: the page ages that value against the server's clock and prints it as the last update, so a report whose group_time is overridden to milliseconds or nanoseconds showed LIVE for ever and a nonsense date. Restart weewxd after installing: the page reads the pinned key, which the service writes only once it has read the new declaration. Declare any time a page does arithmetic on the same way.

8.0.3 2026/10/08

  • long_form() renders on every WeeWX loopdata supports. On WeeWX 4.6 through 4.10.1 every long_form() field was omitted from loop-data.txt -- the sample report's almanac.sun.visible.long_form() among them, so its UV and radiation cards showed no length of day -- because WeeWX's long_form is missing before 4.10 and takes no None_string before 4.10.2. LoopData now composes it itself there, and it reads as on a newer WeeWX ("11 hours, 30 minutes, 36 seconds"). Before 4.10, which has no [[DeltaTimeFormats]] and formats the almanac's and $station.uptime's durations from [[TimeFormats]] as short_delta, brief_delta and long_delta (4.10 renamed them day, hour and month), LoopData formats those three from the report's [[TimeFormats]] entries, which those versions always ship, as the same value's .formatted does; a [[DeltaTimeFormats]] entry under one of those three names, if you add one, comes first.
  • An almanac field naming a body no almanac knows is omitted, with the "Cannot evaluate almanac field" line in the log, on every WeeWX. On WeeWX 4.6 through 5.2 it was published as <weewx.almanac.AlmanacBinder object at 0x...>.
  • The sample report's radiation card says how much shorter or longer the day is than yesterday ("2 minutes shorter than yesterday", "1 minute longer", "under a minute shorter"), where it said "minutes" twice: the field it declared kept WeeWX's unit label. It now declares almanac.sun.visible_change.minute.raw and words the change itself, which also brings it back on WeeWX before 4.10, where the old field failed. On WeeWX before 5.0, which has no almanac.sun.altitude or azimuth, the UV card now gives the sun's transit and sunset, and the radiation card its transit, sunset and end of civil twilight, which were left out with the sentences saying where the sun is; the transit and civil twilight need PyEphem, as they always have. Those sentences still need WeeWX 5.0.
  • At high latitudes, the sample report's UV and radiation cards say in words that the sun does not set today, or does not rise, that civil twilight lasts all night, and that the sun is up "all day", where they put WeeWX's "N/A" into the sentence: "it sets at N/A and civil twilight ends at N/A". The sample skin declares the .raw of the sunrise, the sunset, the end of civil twilight and the day's length beside them, since a time WeeWX has no answer for is published only formatted. Where the sun sets after midnight (Reykjavik for weeks around midsummer), WeeWX's day length comes out negative and reads as the night's length, and its change from yesterday is off by a day when the sunset crosses midnight; the cards now leave both out on those days. Restart weewxd after installing: the service reads the declaration only when it starts, and until then the sun cards leave out the sunrise and sunset, the Sun up tile, the end of civil twilight and the change in daylight from yesterday.
  • The almanac page of the manual gave almanac.sun.az.format(...) and almanac.sun.az.round(1).raw as examples, and az is a plain number, so both were refused. They now use almanac.sun.azimuth, and the page says that azimuth and altitude need WeeWX 5.0.

8.0.2 2026/10/07

  • The sample report says more plainly that its gauges open. The line under the head, which said "Click any gauge for more." in small gray type, is now a band in ink with a tap icon, and says what a card holds: "Click any gauge to see what its reading means, today and on record." On a touch screen it says Tap. The new sentence is translated in all eight shipped languages.

8.0.1 2026/09/28

  • A day aggregate of an observation with no daily summary -- an xtype such as weewx-purple's pm2_5_aqi -- is now refused at startup with the same INFO line every other period has always logged for it, and the field is omitted from loop-data.txt. It used to be produced from the loop packets alone, so day.pm2_5_aqi.min after a restart was the low since the restart, with a time to match, and nothing in the file said so. The rule is now the same for every period: an aggregate needs the observation in the archive (a daily summary), and current. works for anything in the loop packet. A station on the classic wview schema, which archives neither appTemp nor windrun, loses day.appTemp and day.windrun; both were since-restart values there. A period rollover now feeds the new accumulator the pruned packet, as the first one was fed, so a refused observation cannot slip into a period at midnight.

8.0 2026/09/28

  • The sample report's gauges open cards. A click or a tap on any gauge opens a card over the panel (a full-screen sheet on a phone) that says what the reading means and where it sits: the dial and its lines, a sentence -- the change over the trend window, the Beaufort force, the dew point in words, the sun's altitude -- three tiles for today's high and low with their times, and a ladder of the last hour, the last 24 hours, today, this week, this month, this year and all time, each a bar from the period's low to its high with the reading now drawn through every row and the time or date of each extreme under it. The windrose card draws the rose for this hour, today, this week, this month and this year; the UV and air quality cards color their peaks in EPA's categories and show EPA's scale with the reading's row marked, the air quality card ranking each period's highest, and today's lowest, PM2.5 as the index weewx-purple computes for it, since the index is never archived and an aggregate of it would count only the packets since weewxd started. The cards step with the arrow keys or the buttons in their head, close on Escape, the x or a click outside, and follow the station on every packet while open. Everything they say comes from loop-data.txt: the skin declares the fields, one group per card, in its skin.conf. The new rolling 24-hour windows cost the service about 10 MB each at two-second packets, some 80 MB for the nine observations the cards rank.
  • The panel's dew point gauge says the dew point in words -- dry, comfortable, sticky, muggy, oppressive -- where its word line was blank.
  • Every card string is translated in the eight shipped languages. The translations have not been reviewed by native speakers.

7.5.2 2026/09/26

  • The sample report's wind compass points with the same red arm as the dials beside it, and the gust with a plain lighter needle; both used to carry arrowheads. The tip marks where the wind comes from, and an arrowhead invited reading it the other way, as the way the wind is going. In a calm, with no red arm to carry, the hub takes the gust needle's color.
  • The barometer's trend arc is drawn only when the change is big enough to show its head, about 1.3 hPa (0.04 inHg) over the trend window. A smaller change used to draw a sliver too small to see or to read as an arrow.

7.5.1 2026/09/23

  • The sample report's windrose sits on a dial face of its own, rimmed like the gauges beside it with the compass letters inside, and that face is exactly the size of theirs; in 7.5 the rose was drawn straight on its card and came out a different size from the dials in its row. Its letters and its lines -- the rim, the rings, the hub's edge -- are drawn the same size on screen as the dials' too. Its rings are a shade darker in the light theme, so they still stand out against the face.

7.5 2026/09/23

  • The sample report's gauges are redrawn. Each is a card: the dial, then under it the reading, a line saying what it means, and today's high above its low. The light theme now draws light dials -- a pale face on a white card -- where it used to put the dark dials on a light page, and the dark theme draws deep dials on navy cards. Each gauge has one red arm, as an analog gauge does; the wind's gust needle keeps a quieter color of its own. The numerals stay inside the face, where they were, but no reading or caption is written across a dial any more; each sits below its dial instead.
  • UV and air quality wear EPA's category colors on the rim and name the category under the reading. UV follows EPA in naming the index rounded, so 5.5 is High. The air quality dial now runs to 500, the top of EPA's Hazardous band, where it used to stop at 300.
  • The wind gauge says which way the wind is from and how hard it is gusting, gives today's peak, and marks today's prevailing direction on the dial. Feels like says how much warmer or cooler it is than the air, and rain says when the day has been dry so far.
  • The numerals on the dials use the report's decimal point, as the readings under them already did.
  • Fixed: on a metric station the rain and rain-rate dials printed their major numerals rounded -- "6", "13", "19" beside ticks at 6.25, 12.5 and 18.75 mm. The metric scale now runs to 20 mm, so every major tick falls on a whole millimeter, and so does every doubling on a big day.
  • The gauges are SVG rather than canvas, and every color is a css property in index.html.tmpl, including the windrose's six speed-band shades, which were the last colors written in the javascript.
  • The sample report declares more fields in its skin.conf, for the readings, units, peaks and wind words above. A station that overrides one of the sample report's field groups under its stanza in weewx.conf replaces that group whole, so a gauge whose group was overridden shows "--" for any field the override leaves out; "What each gauge reads" in the manual lists what each gauge needs.
  • Nineteen new strings for the gauges, in all nine languages. The translations of those strings have not yet been reviewed by native speakers, German, French and Danish included.
  • Internal: the change history is now changes.md, in Markdown, so it reads as formatted text on GitHub; it was changes.txt.

7.4.1 2026/09/17

  • Fixed: on WeeWX earlier than 5.3, every report that declares a field holding a time lost its whole entry in loop-data.txt, on every packet, with one error logged for each such report. 7.4 reads the default time format -- the one a report uses for a time context its [[TimeFormats]] does not name -- from weewx.units.DEFAULT_TIME_FORMAT, and WeeWX added that name in 5.3; 4.6 through 5.2 write the same format inline and have no such name to read. It is now resolved once, at startup, and where the name does not exist the format those releases write is used, so a time renders exactly as it does on the report's own pages. A station running WeeWX 5.3 or later was never affected, in 7.4 or now. Reported by bellrichm (issue #17), citing WeeWX 5.0.0.
  • Internal: three tests now ask the running WeeWX what it does rather than assuming the newest release's behavior, so the whole suite runs unchanged against an older WeeWX. That is what guards the fix above: it is checked against the oldest WeeWX the suite can run on, not only against the one this machine happens to have.

7.4 2026/09/10

  • Times and numbers now render in the language of the report that declared them, as everything else already did. A month or weekday name -- 21. Juni, samedi -- and the decimal point in a formatted number come from the locale, and WeeWX keeps the locale in one process-wide setting that it changes for the duration of each report it generates. LoopData writes its file on every loop packet, in a thread of its own, so it was reading whatever that setting happened to hold: a packet that landed while some report was being generated took that report's language and the next packet took weewxd's, and the same field could come out two ways one packet apart. Each report's locale is now read once, at startup, and kept with its formatter and converter, so every value a report declares is rendered in that report's own language whatever else weewxd is doing. This covers station.rain_year_str too, which WeeWX fixes when it builds the station rather than when the field is read. A station whose reports do not set lang, or whose lang is not a locale name the machine has installed, sees no change at all.
  • A formatted number now carries the report's decimal point, as the report's own pages always have. On a station running in a language that writes 12,3 rather than 12.3, a field's .formatted was the one rendering that skipped this, so the page said 12,3 and .formatted beside it in loop-data.txt said 12.3 -- the same number, in the same file, for the same packet. Unformatted .raw values are untouched: they are json numbers, and always were.
  • Internal: two errors LoopData can log were missing from the manual's table of log messages, one of them because the test that keeps that table honest could not see a message written with double quotes.

7.3 2026/09/09

  • ACTION REQUIRED IF YOU WANT THE DARK PAGE ON EVERY SCREEN: the sample report now follows the viewer's own system theme by default, and a browser with no preference set reports LIGHT rather than "no preference" -- so a machine where nobody has chosen a theme now gets the light page where it used to get the dark one. To pin the dark page for everyone regardless of their setting, add theme = dark to the report's [[[Extras]]] in weewx.conf. A fresh install writes the option commented out; an existing install has no such line and needs one added. The manual's "Themes" section spells it out.
  • The sample report has a light theme -- deep navy instruments on warm paper -- and picks between it and the dark one from the viewer's system setting. Two people looking at the same page can see different themes, and a viewer who changes their setting while the page is open sees it change without a reload. theme = auto (the default), dark or light in the report's [[[Extras]]] decides; anything unrecognized is treated as auto. The page also declares its color-scheme now, so the browser's own scrollbars and focus rings follow the theme instead of staying light on a dark page.
  • The sample page's gauges sit on rounded panels, and every row of the grid now reserves the height of the windrose's legend rather than only the row the windrose is in. THIS ONE MOVES THE DARK PAGE TOO: the bottom row of gauges sits about twenty pixels lower than it did in 7.2, so that the dials in each row line up with each other rather than the first row standing taller than the rest.
  • Internal: the panel's palette is now written ONCE, as the css custom properties in index.html.tmpl. realtime_updater.inc reads them rather than carrying a second copy of every color, so a theme is a css change and nothing else, and the test that kept the two lists in step is gone with the duplication it existed for. The needle's glow, which was a third copy of the amber spelled rgba(), is derived from it.

7.2 2026/09/06

  • A period's own boundaries are now fields. start, end, length and dateTime are report tags on every period ($week.start, $month.length), and loopdata refused all of them. Declare week.start.raw and every packet carries the instant this week began, so a page can work out for itself whether a period began today instead of being told once, at report time, by an answer that goes stale the moment midnight passes -- which is what a kiosk page left up for a fortnight is living with. All four are served on hour, day, week, month, year, rainyear and alltime, and take the usual unit override, round(n) and format specs. start, end and dateTime are instants and render through the period's time format; length is a duration and renders as one. The rolling periods (1m-1440m, 1h-24h) and trend have no fixed span and do not carry them.
  • alltime.start is the time of the earliest record in the archive and alltime.end the time of the packet in hand, which is what the $alltime report tag spans. alltime.end and alltime.length therefore move with every packet, where the calendar periods' do not. That earliest record is read from the database once, when the accumulators are built on the first loop packet, and thereafter kept current from the archive records themselves, at no further cost to the database (except where [StdArchive] writes a different data_binding than the reports read, in which case loopdata cannot tell which archive a record belongs to and keeps to the startup reading, saying so in the log). So a station whose archive was still empty at that moment -- weewx and loopdata installed together -- serves the alltime span properties from its first archive record, five minutes in by default, rather than waiting for a restart. A packet stamped before the archive's earliest record, which a console with a reset clock can produce, has no alltime span at all, so those fields are omitted for that packet. Neither case affects any other field.
  • A period's dateTime with an aggregate on it -- day.dateTime.max and its kin -- is no longer recognized, since a period's dateTime is now its start. This is what the report tags have always done: $day.dateTime is the day's start, never the dateTime observation. current.dateTime.raw is untouched; current is an instant, not a span. The same holds for the other three names: under a period that has a span, start, end and length are the span's, so an observation named one of them -- day.length.max -- is no longer recognized there either. No standard weewx type carries any of those names.
  • Internal: six tests drive the span properties through the machinery the first round stubbed out -- a period rolling over on a real accumulator, with the observations declared that make it roll at all; an archive record's stamp arriving among loop packets on the live writer thread, with every file written checked rather than the last one left on disk; a report declaring span properties and nothing else, end to end through a real service; the deprecated [[Include]] fields line, where a property the target report also declares is rendered once and copied flat; two reports in different unit systems reading one span; and the two days a year that are not twenty-four hours long. The report-tag parity test now compares the rendered string as well as the raw instant, which is where a wrong time context shows.

7.1 2026/09/03

  • Week, month, year, rain-year and all-time sums, counts and averages no longer count today twice after a restart. Loopdata seeds those spans from the daily summary table and then adds today's running totals on top, but today has a row in that table too, so from a restart on, month.rain.sum and its kin read today's contribution double and every span average leaned toward today's mean. Minimums and maximums were unaffected. The error lasted until the span rolled over -- and for the all-time span, which never does, until the next restart doubled it afresh.
  • When weewxd stops, loopdata stops its writer thread before the process ends, so a stop that lands while a packet is being written does not leave a LoopData temp file behind in the directory the loop-data file is served from. Packets still queued at that moment are discarded rather than written and shipped to nobody. Should the thread not stop within twenty seconds -- a stalled rsync could hold it that long -- the stop goes ahead without it and says so in the log. A driver restart, where weewxd rebuilds its engine in the same process, likewise no longer leaves the previous writer thread and its accumulators behind.
  • Should the writer thread die -- its traceback has always been logged -- loop packets are no longer queued to it for the rest of weewxd's life, which grew memory by one packet every loop interval while the loop-data file sat frozen. The loss is logged once more, saying the file will not update again.
  • [[RsyncSpec]] skip_if_older_than must be at least 1. Zero used to mean never skip, undocumented, and a station that set it would rsync every packet however late -- the backlog the option exists to prevent. A value below 1, and a blank one, are now ignored with a warning and the default of 3 is used.
  • When weewxd starts, the day accumulator is seeded from the daily summary of the first loop packet's day rather than the wall clock's. The two differ only when a start straddles midnight in the second or so between a packet being stamped and being processed -- exceedingly unlikely in practice -- but when they did, the day's min/max/sum carried the wrong day's history until the packet-time day rolled over.
  • Internal: the two archive database connections loopdata opened -- one when the service starts, one when the first loop packet builds the accumulators -- are closed when they are finished with, rather than held open for the life of weewxd.
  • Internal: the test suite closes the files it writes, so a run no longer prints a ResourceWarning for each one.

7.0.1 2026/08/31

  • Nothing below edits a weewx.conf that already exists, and there is nothing to do by hand. The one change that reaches a station without an edit is the expiration_time default, immediately below.
  • The sample page's default expiration_time is 24 hours, where it was 4, matching the sibling extensions. This reaches a station that has no live expiration_time in weewx.conf: a fresh install, where the option is written commented out. A station whose weewx.conf carries expiration_time = 4 as a live setting -- every install made before the option was demoted -- goes on expiring after 4 hours until that line is edited or removed.
  • A fresh install writes eight options into weewx.conf commented out, each showing the value that will apply: compress, log_success, timeout and skip_if_older_than in [[RsyncSpec]], and loop_data_file, expiration_time, googleAnalyticsId and analytics_host in the sample report's [Extras] (the last two as examples -- their default is to be absent). weectl fills in absent keys and never rewrites one, so an option written live freezes that station on today's default for ever, while a commented one leaves loopdata's own value to answer, including a better one a later release brings. What stays live is what loopdata cannot supply for itself: the switches (enable), the placeholders a user must replace (remote_server, remote_user, remote_dir, page_update_pwd), what weectl needs (HTML_ROOT, skin), the station's loop cadence, the file's directory and name, and the sample report's string formats -- which are on the report's stanza so that a station-wide [[Defaults]] cannot put a decimal on a wind dial that has no room for one.
  • Fixed: deleting enable, compress or log_success from [[RsyncSpec]] -- or deleting the whole section -- stopped LoopData at startup with "Unknown boolean specifier: 'None'". Each now falls back to false, which is what the installer has always written.
  • The installer's [[ScriptData]] example of how to declare fields no longer lands in weewx.conf indented as though it belonged to [StdReport]. ConfigObj attaches a comment block to the next key and writes it at that key's indent, and the block sat after the section's last option.
  • Fixed: a kiosk URL whose ?pageUpdate= password contains a space, or any other character a browser percent-encodes, now matches. The password was compared still encoded, so such a password never exempted the page and the display went dead after expiration_time hours. The URL value is compared both as typed and percent-decoded, so a URL that worked before -- including a password that itself contains a percent-escape -- goes on working untouched.
  • Fixed: a quote or a backslash in page_update_pwd, refresh_rate or expiration_time no longer kills the page outright, and a page_update_pwd containing an accent or any other non-ASCII character now matches. All three were written into the javascript as hand-quoted literals: a quote ended the string early, which is a parse error that stops the whole script -- no polling, no gauges -- and a non-ASCII password was rendered as a character reference, which could never equal what the URL supplied. All three are emitted with json.dumps now, as the page's translated strings already were, so a mistyped number falls back to its default instead of taking the page down.
  • Fixed: blanking page_update_pwd no longer leaves every visitor's page polling for ever, and no URL exempts a page whose password is blank. An empty value used to be read as a password, so the empty ?pageUpdate= every visitor has matched it; it then fell back to a fixed word that is in the shipped source for anyone to read. Now there is no password at all, which nothing in a URL can equal. (An [Extras] option written empty is present, so the test for one fired on it.)
  • expiration_time = 0 now means the page never expires, which is what a zero timeout ought to mean and what it never did: it expired the page before the first packet arrived, since ConfigObj yields strings and a '0' passed the template's test for a value. It is the kiosk exemption without the ?pageUpdate= URL.
  • Fixed: refresh_rate = 0 no longer polls as fast as the browser can; a page that never polls is a dead page, so zero means the default there. A negative, or a setting that is not a number, means the default for both -- rather than emitting javascript that stopped the whole script, and, for a negative expiration_time, rather than quietly never expiring.
  • Fixed: a large expiration_time no longer expires the page at once. Past 2147483647 milliseconds (~596 hours) the delay overflows the browser's int32 timer and fires immediately, so an expiration_time of 1000, reached for as "effectively never", did the opposite. It is clamped to that ceiling now, and zero is how you say never.
  • Fixed: once the page has expired, the CLICK-ME prompt is written once rather than rewritten every couple of seconds for as long as the tab is open. The test meant to prevent that compared the element itself against the string.
  • Fixed: a page adapted without the sample skin's status elements no longer throws on every poll. Each of the five places that writes the LIVE indicator or the timestamp -- including the OFFLINE, NO DATA and BAD DATA branches -- now goes through one guarded writer, and the gauges are drawn before the labels rather than after, so a missing label cannot stop the drawing.
  • Fixed: a page adapted without the windrose legend's element no longer takes the eleven dials down with it. The lookup is guarded, and the legend's cache key is now set only once the legend is built, which is what made the symptom intermittent.
  • Fixed: a report that declares gauge fields but not the clock group now says BAD DATA -- check loop_data_file instead of reading "NaNs ago" and "Invalid Date" for ever. The gauges still draw.
  • Fixed: the sample page's windrose colors stretch to however many speed bands are configured. The six shades were indexed one per band, so a seventh band wrapped around and painted the windiest band in the calmest band's color -- in the rose and in its legend alike, so the page did not even look wrong. They are stops on a curve now, sampled for the band count in use; six bands, which is what the default edges and the shipped sample configuration both produce, get exactly the six shades they always did.
  • A README correction: the list of [LoopData] options had every rsync option, plus the loop frequency, sitting under the heading that marks the three deprecated ones -- so options a fresh install does write, and the block just above it shows being written, read as deprecated. The deprecated three (target_report, fields and the [LoopData] spelling of windrose_bands) are together at the end of the list now, under that heading and nothing else.
  • Fixed: the sample page's data age is measured against the web server's clock instead of the browser's, so a viewer whose device has the wrong time no longer sees a false age -- or, since the page reads LIVE only while the packet is younger than three times refresh_rate (six seconds by default), never sees LIVE at all over a perfectly healthy feed. The age now comes from the Date header of the response that carried the record, plus any Age a proxy reports, and never reads the viewer's clock while that header is there. A monotonic floor -- how long since the data last changed -- catches a cached copy whose header reads younger than its body. Where no Date header reaches the page at all the viewer's clock is still the only measure available and is used, since otherwise a feed that had already stopped before the page loaded would read LIVE.
  • Fixed: running weectl report run no longer drops a zero-byte LoopData file into the directory the loop-data file is written to -- which is inside the web-served report tree by default, and got one more on every run. LoopData reserves a unique name at startup to write each packet to before renaming it onto loop-data.txt, and was leaving the empty file itself behind; only the name was wanted. weewxd never left one, which is why this went unseen: the file is renamed away by every packet, and the cleanup that existed ran on the loop-processor thread, which weectl report run never starts even though report_services builds the service.
  • The windrose manual page gains a picture of one: the sample skin's rose on a real morning, with the default bands, captioned so that the calm center and the two-lobed shape illustrate the advice on choosing edges that follows it.
  • Manual corrections: the windrose page's advice on the top band edge said the darkest color flags the strongest wind. It is the other way round on the sample page, whose ramp runs dim for calm to bright for windy; some other windrose tools do run dark for fast, which is where the sentence came from. The troubleshooting page's BAD DATA row gave only one of the three things that show it -- a body that is not json -- and not the two that name a report: no entry for this report at all, or an entry with no current.dateTime.raw in it.
  • Fixed: a gauge whose cell is hidden because the station has no such sensor is re-sized when the sensor reappears. A hidden element measures zero, so its canvas was built at a fallback size, and only the windrose was ever checked for that drift.
  • Internal: three guard tests on the installer stanza -- the commented options pinned as a complete set with their values, each value checked against the fallback that governs when the key is absent, and comment placement checked through weecfg's own merge, counted so that a block dropped outright (which happens silently when it attaches to a section weewx.conf already has) fails.

7.0 2026/08/26

  • ACTION REQUIRED, 1 of 3 -- WeeWX 4.6 or later: this release requires WeeWX 4.6 or later, where 6.x required 4.0. Both the installer and the service refuse to run on anything older. 4.6 is where WeeWX made its skin-dictionary builder a module-level function, which is how loopdata assembles each report's units, formats and language, and it is also where $gettext arrived, which the sample skin has needed since 6.4.
  • ACTION REQUIRED, 2 of 3 -- FINISH THE MIGRATION by running user.loopdata as a command (the manual's Declaring fields page gives the command line for each way of installing WeeWX). The fields line does not remove itself, and an upgraded station should remove it once every extension whose pages read the loop-data file declares its own fields. The command reports which report now covers each entry on the line and which are covered by nobody -- almost always cruft, entries a page asked for years ago and no longer reads -- and changes nothing until every entry is accounted for and it is run with --apply, which backs up weewx.conf, deletes [[Include]] and [[Formatting]], and moves [LoopData] windrose_bands onto the stanza of the report whose rose it bands (see the windrose_bands bullet below). A relative loop_data_dir is pinned to the directory the file is in today first, since it was measured from target_report and that is going -- no page finds its URL gone. It also runs with the fields line already deleted, to remove what is left. The startup warning names the command.
  • ACTION REQUIRED, 3 of 3 -- a live page of your own that reads loop-data.txt: declare its fields under its report and have the page read that report's entry, before the fields line goes. Nothing is required for the sample page, weewx-celestial or weewx-weatherboard: upgrade each and it declares its own fields, and until then they keep working through the fields line exactly as before.
  • Reports declare the fields they need. A skin's skin.conf carries a [LoopData] [[fields]] section of named groups (see skins/LoopData/skin.conf), and every enabled report whose skin declares fields is its own target: its fields are rendered with its own formatter, converter, [Texts], [Almanac] names, $station and [Units] [Trend] time_delta, and written into loop-data.txt under the REPORT's name (the [StdReport] section name -- one skin can be listed under two reports, in two unit systems, and each gets its own values). A report can add to or override its skin's groups from weewx.conf ([StdReport] [[MyReport]] [[[LoopData]]] [[[[fields]]]]), which is how an extension declares fields that depend on your configuration. Accumulation is shared: one set of accumulators feeds every report. The two report settings that reach the accumulators -- the trend window and the windrose band edges -- select an accumulator by value, so reports that agree share one and a report that differs gets its own.
  • A fields line entry that any report declares and renders identically is rendered ONCE, in that report's entry, and copied flat, so a station still carrying the line pays almost nothing for it. Two reports count as rendering identically when their units, formats, labels, texts, almanac names, trend window and windrose bands all match, which on a station whose reports inherit [StdReport] [[Defaults]] is every report. Only fields nothing declares, or that no matching report declares, are rendered for the line itself.
  • windrose_bands is a report option. On a report's stanza in [StdReport] it is in that report's windSpeed unit; under [StdReport] [[Defaults]] it applies to every report, in the Defaults' unit, converted to each report's and rounded to one decimal; a skin.conf may carry one too. The pre-7.0 [LoopData] windrose_bands is deprecated. It bands target_report's rose and no other -- which is the one rose it banded before 7.0, the one in the flat file -- so that report keeps its bands and every other report takes the WRPLOT defaults until someone chooses otherwise. Finishing the migration moves the value to that report's stanza, in the unit it is already written in, or simply drops it when that report already gets its bands from somewhere nearer.
  • The [LoopData] [[Include]] fields line and [[Formatting]] target_report are deprecated, together with the flat top-level keys they produce. All three still work exactly as before on an upgraded station, so an existing page keeps reading what it always read, and the per-report entries appear in the same file beside them. A fresh install writes neither. weewxd logs a warning at startup while the line is present; do not edit it by hand -- an extension declares its own fields when installed, and a later loopdata release removes the line once every extension that used it declares its fields. A live page of your own should declare its fields under its report (see the manual's Declaring fields page) before that happens.
  • target_report keeps one job while it exists: a relative loop_data_dir is relative to its directory, LoopDataReport when target_report is not set. So keep the [[LoopDataReport]] section -- enable = false turns the page off, and the installer puts a deleted section back. A target_report kept after the fields line is gone is warned about at startup: set loop_data_dir to an absolute path before the release that removes it, or the file moves.
  • A FRESH install writes no fields line, and weewx-celestial 8.4's installer only adds to a line that already exists, so a fresh station needs a weewx-celestial newer than 8.4 for the Celestial page to have live values. Upgraded stations keep their line and are unaffected.
  • One report's fields failing to render no longer stops the file. Every enabled report's declaration is rendered on every packet now, including skins this station's operator never wrote; a failure drops that report's entry, logs once and leaves every other report's values written, where before it would have ended the loop-data thread and frozen the file while weewxd carried on.
  • A report for consumers that are not reports. A script, an SNMP extension or a monitoring check has no skin to declare fields in, and parking them on a page's report means an upgrade overwrites them. The installer now writes a disabled [StdReport] [[ScriptData]] stanza and ships skins/ScriptData, a skin that generates nothing: enable the report, declare your fields under its stanza in weewx.conf, and they arrive in the loop-data file under "ScriptData". The manual's Declaring fields page has the details.
  • A field none of the parsers accept is now logged as a warning, naming the report that declared it. It was dropped silently.
  • A single-field fields value (no comma, which ConfigObj hands back as a string) is read as one field. It was iterated character by character.
  • The sample skin declares its fields in its skin.conf, grouped by gauge, and its page reads its own report's entries. The installer no longer writes a fields line or a target_report.
  • Internal: Configuration is split into the shared half and a ReportContext per report; generate_loopdata_dictionary is split into accumulate_packet and render_report; the almanac and station evaluators are per report (their caches hold rendered values).

6.11.3 2026/08/24

  • The sample page no longer loads Google Analytics when no analytics id is configured. analytics.inc tested whether googleAnalyticsId was PRESENT, not whether it had a value -- and the installer writes it present but empty. Every install that took the default therefore fetched https://www.googletagmanager.com/gtag/js?id= on every page view, with an empty id: a third-party request nobody asked for, on a page whose owner had never configured analytics. The companion option failed the same way in the other direction: with analytics_host present but empty the page compared its hostname against "", which is never true, so a user who did set an id and left the host alone got the script loaded and not one event reported. Both now test the value. Empty means off for googleAnalyticsId, and means report from wherever the page is served for analytics_host -- which is what both options always claimed. Existing installs pick this up on their next report cycle; no configuration change is needed. The manual documents both options for the first time -- they had never appeared in docs/ or the README, and the test that is supposed to catch that read only skin.conf, where both sit commented out and so were invisible to it. It now also reads the [[[Extras]]] the installer writes.
  • Blanking refresh_rate or expiration_time in weewx.conf no longer kills the sample page. realtime_updater.inc tested those two for PRESENCE and emitted the value straight into a javascript assignment, so an empty one rendered "refresh_rate = ;" -- a parse error that took down the whole script block, and with it the polling, the gauges and the translated strings, not merely the one setting. Both now test the value and fall back to their defaults, as a missing option always did. Same fix as the analytics options above; page_update_pwd was never affected, its value being quoted.
  • The [LoopData] stanza a fresh install writes into weewx.conf arrives commented. It had always landed as bare key = value lines with nothing to say what any of them did, which left the manual as the only way to find out what, say, [[LoopFrequency]] seconds was for -- and that is the option whose wrong value silently scales every windrun and every windrose distance. Now each section and each option carries a line or two explaining itself, the three placeholders in [[RsyncSpec]] say they are placeholders, and the fields line says what a field is and where the full grammar lives. The cause was that the installer built its stanza as a Python dict: weectl merges it with weeutil.config.conditional_merge, which copies a key's comments across only if the source has any, and a dict has none. The stanza is now a ConfigObj, which does. Existing installs are untouched -- the merge fills in absent keys only, and every value written is unchanged, so this matters only on a fresh install.
  • The manual no longer says the sample panel calls its data stale after expiration_time seconds. Both halves were wrong: expiration_time is in hours, and it is the poll-forever timeout, not a staleness test. The panel reads LIVE while the packet is no older than three times the skin's refresh_rate -- six seconds by default -- and shows the age instead once it is older. The passage's point is unchanged; only its numbers were.
  • Internal: three tests pin the new stanza -- that every section and option in it is commented, that the comments survive a merge driven through weecfg's own ExtensionEngine._inject_config (which is where the stanza is copied, and the copy is what decides whether comments survive at all), and the values a fresh install writes.
  • Internal: install.py builds its ConfigObj inside the installer rather than at import time, so loader()'s "requires WeeWX 4" message is reachable again. A module-level import of weeutil.config ran before that check and could not survive a WeeWX 3 install -- the module is absent before 3.9, and 3.9's copy has no config_from_str -- so the user got a raw traceback instead of the message.

6.11.2 2026/08/24

  • The manual says where the loop-data file should live. The default -- inside your reports tree, in the target report's own directory -- works, and is what most stations keep. Two things are worth knowing all the same: the file is rewritten on every loop packet, which a station running from an SD card may rather have land in RAM, and the report sync copies that tree every report cycle, which buys nothing when the pages are served from the same machine. A memory filesystem outside the web root answers both, for anyone comfortable editing a web server configuration: the new section carries the /dev/shm and fstab recipes, the Apache and nginx aliases (with the no-store header a live page wants), the URL the reading page then needs, and the case none of it helps: where the report sync is the only way to reach the machine serving the pages, as it is when publishing by FTP, the file crosses only at report cadence and cannot feed a live page at all. That wants loopdata's own rsync, or pages served from the machine weewx runs on.
  • Upgrading gains a 6.11.2 line.
  • The manual no longer calls the loop frequency crucial without saying what it costs. It is crucial; now it says why. Only live packets are weighted with this value: every period that outlives weewxd's startup -- hour, day, week, month, year, rainyear, alltime and the rolling windows -- is seeded from what is already in the database, at the weights the archive gives it. So a wrong value skews the ratio between the seeded part of a period and the live part rather than canceling out of the average, and the manual works an example through: a station emitting every two seconds but configured for four counts its evening twice over, and day.outTemp.avg reads 16.0 where it should read 17.5. The skew is bounded by the period's own length -- it clears when the period rolls or the seeded records age out of a rolling window -- and minima and maxima never use the weight at all. windrun and windrose are the ones that do not recover: both are wind speed times that interval, so they scale straight with the error and no roll-over cleans them, and day.windrose.banded is what the sample panel draws. Two seconds is also no longer stated as a fact about hardware in general: it is what a Davis Vantage emits, with nothing to set on the station side, while a driver you poll on a schedule hands you the number directly.
  • The manual carries the new section everywhere a reader chooses the value, not just where the option is defined: Installation and Build a live page both point at it, and Troubleshooting's NO DATA (HTTP 404) row no longer reads as a warning against the very arrangement the manual now recommends -- keeping the file outside the web root is fine, it just has to be served and named.
  • The fstab recipe in that section says to create the mount point and give it to the user weewxd runs as. A root-owned mode=0755 tmpfs is not writable by a weewxd running as anyone else, and the failure is unhelpful: the directory exists so makedirs succeeds, and it is the temp file opened next that raises PermissionError, during service init, so weewxd does not start. /dev/shm needs neither step.
  • Internal: the absolute-manual-URL audit checks a link's #fragment against the target page's headings again. It had been made to strip the fragment so an anchored link would pass the .html check, which left the repo's first anchored absolute link unverified -- rename the section and the link would have died silently.
  • The sample skin's shipped loop_data_file is the value every install actually runs: loop-data.txt, the file beside the page, matching the default loop_data_dir = . . skin.conf shipped ../loop-data.txt -- dead text, since weewx.conf's own entry overrides it -- and the manual documented that value and explained it wrongly. No install changes behavior; a reader of either file is no longer misled.

6.11.1 2026/08/14

  • The sample report drops the loopdata field names printed under each gauge. They were meant as a key to what the panel reads, but on a live page they read as a template that failed to render -- one user asked which part of his install was broken. The mapping lives in the manual instead, gauge by gauge.
  • Troubleshooting covers a single gauge missing its needle. An old fields line carried over an upgrade is already documented, but only as the whole panel losing its needles, bands and windrose; a hand-tuned fields line can be short one .raw field and cost exactly one needle, which is what the same user hit before he replaced his fields line.

6.11 2026/08/13

  • The sample report's dark theme is higher contrast and much easier to read: white readouts, brighter numerals and labels, a recessed track you can see on every dial, and windrose bands that no longer vanish into the disc. The field captions lose an opacity that had them under AA for text that size.
  • The windrose is now the same size as the other eleven gauges, with its N/E/S/W letters moved inside the disc.
  • Internal: a test pins the panel's two palettes to each other -- :root in index.html.tmpl and the C object in realtime_updater.inc -- since the canvases cannot read css variables and nothing else stopped the copies drifting.

6.10 2026/08/13

  • Almanac fields work again on WeeWX earlier than 5.3. WeeWX 5.3 renamed the Almanac constructor's language argument: it now takes texts (the target report's whole [Almanac] section -- moon phases, body names, constellation names), where earlier versions took moon_phases, the eight phase names alone. Since almanac fields shipped in 5.0 loopdata has passed texts unconditionally, which is a TypeError on WeeWX 4.x through 5.2: at the first loop packet the LoopData thread logged a traceback and stopped, and loop-data.txt was never written again -- weewxd itself kept running, so the only signs were the traceback in the log and a page frozen at its last values. A fields line with no almanac field never built an Almanac and was unaffected. loopdata now asks the Almanac constructor which spelling it takes and passes that one; on a pre-5.3 WeeWX it passes the moon phases when the target report overrides them and otherwise leaves WeeWX's own default in place -- what a report on that version does. Reported by bellrichm (issue #15), running 5.2.
  • The manual and README have been overhauled. The manual gains an Upgrading page that gathers the action-required items previously scattered across four pages, and the README is now a front door that deep-links into the manual rather than duplicating it. The [[RsyncSpec]] remote_port option is documented for the first time -- the code has always read it -- and so is [Extras] refresh_rate. The driver-support note now says what a mismatched driver actually costs (windrun and the other time integrals, not the averages) instead of "at your own risk". The two items below came out of that review.
  • The German and French translations are marked Beta again. The manual described both as native-speaker reviewed, and 6.6 declared the French out of Beta, on the strength of reviews that covered the vocabulary this skin shares with weewx-skyfield and weewx-celestial -- body names, moon phases, compass ordinates and the constellation names -- which is copied verbatim from those files. Those reviews did not cover the strings only this skin renders: of the 36 [Texts] keys in lang/de.conf, 7 are shared and 29 are loopdata's own, including every gauge heading and all nine trend.barometer.desc descriptions. The translations themselves are unchanged; only the claim made about them is. Corrections welcome.
  • The sample skin's skin.conf comments are corrected. The URL parameter that keeps a page auto-updating is pageUpdate, not page_update_pwd as the comment had said since the sample skin shipped in 2.6 -- a reader who followed the comment would have had the page time out on them. The lang comment, four releases stale, now lists all eight translations, and two typos are fixed.

6.9 2026/08/08

  • A next_* almanac field now expires the moment its event's own instant passes, instead of being kept until midnight: a "next" value whose instant is behind us no longer is what the field name promises. Fields naming the same event -- say a satellite pass's rise, set and maximum altitude -- are computed together and expire together, at the latest time-typed value among them (for a pass, the set time), so an in-progress pass keeps serving until it ends and the very next packet carries the following pass, every field switching at once. Recomputes stay rare by construction, since each one returns the next occurrence, re-arming the cache: next_full_moon recomputes once per lunar month, a satellite's next_pass once per pass. A group with no time-typed field cannot see its event end and keeps the once-per-day refresh -- include a pass time field to get event-time expiry. previous_* fields are unchanged (their instant is always in the past; they refresh once per local day). The stale window has been there since almanac fields shipped in 5.0 -- almanac.next_full_moon serves the old event from the instant it passes until midnight -- but it took developing weewx-celestial 8.0's "Next pass overhead" panel against this code to make it glaring: driven by the old caching, the panel showed a morning pass as "just set" until the day's end. Fixed before either shipped.
  • A day- or event-tier almanac field that evaluates to no data is no longer cached until the next local day: every packet retries, so the field picks up its value the moment the almanac has data. Seen live with weewx-skyfield satellites: a newly added satellite's rise, set and pass fields honestly read N/A until its orbital elements downloaded, but before this fix they kept reading N/A until midnight even though the elements arrived moments after startup.
  • Complete Italian, Norwegian (Bokmål) and Swedish translations ship with the sample LoopData skin (all three Beta, awaiting native-speaker review -- corrections welcome): lang/it.conf, lang/no.conf and lang/sv.conf, each including the trend.barometer.desc tendency vocabulary and, for a loopdata instance whose target report is in that language, the moon phases, body names and all 88 constellation names -- the shared vocabulary in step with weewx-skyfield's and weewx-celestial's files of the same names (the cross-repo test pins it, as for the other languages). Select one with lang = it (or no, sv) in the report. Matches the three new languages of weewx-skyfield 2.0 and weewx-celestial 8.0.

6.8 2026/08/02

  • A complete Danish translation ships with the sample LoopData skin, contributed by native speaker Gert Andersen: lang/da.conf, including the trend.barometer.desc tendency vocabulary and the shared almanac vocabulary (body names, moon phases, compass points and all 88 constellations) in step with weewx-skyfield's and weewx-celestial's da.conf files (a test pins this). Select it with lang = da in the report.

6.7 2026/07/31

  • A complete Spanish translation ships with the sample LoopData skin (Beta, awaiting native-speaker review -- corrections welcome): lang/es.conf, including the trend.barometer.desc tendency vocabulary and, for a loopdata instance whose target report is Spanish, the moon phases, body names and all 88 constellation names -- the shared vocabulary in step with weewx-skyfield's and weewx-celestial's es.conf (the cross-repo test pins it, as for the German, French and Dutch). Select it with lang = es in the report.

6.6 2026/07/30

  • Almanac fields take a unit segment, completing the grammar -- observation fields (day.outTemp.avg.degree_C) and station fields (station.altitude.meter.raw) already took one: almanac.sunrise.unix_epoch.raw, almanac.sun.visible.second.raw, almanac.sun.visible.hour.round(2).raw. The unit segment pins the unit no matter what the target report's unit settings say. Without one, raw values follow the target report's converter (unchanged, by design) -- including any [Units] [[Groups]] overrides, so a report that sets group_deltatime = hour turns almanac.sun.visible.raw into hours. Pin the unit on any .raw field a page consumes numerically.
  • Attribute chains now walk exactly as Cheetah's NameMapper walks a report tag: a segment that resolves to a method or function is auto-called at every dot, not only at the end of the chain (under weewx-skyfield <= 1.14, almanac.moon.parallactic_angle.degrees failed as a loopdata field even though the report tag worked), and a callable instance is never auto-called, exactly as Cheetah leaves them alone. Applies to almanac and station fields alike.
  • The French translation is out of Beta: lang/fr.conf has been native-speaker reviewed, in step with weewx-skyfield's and weewx-celestial's reviewed French (the cross-repo test pins the three together).
  • A complete Dutch translation ships with the sample LoopData skin (Beta, awaiting native-speaker review -- corrections welcome): lang/nl.conf, including the trend.barometer.desc tendency vocabulary and, for a loopdata instance whose target report is Dutch, the moon phases, body names and all 88 constellation names -- the shared vocabulary in step with weewx-skyfield's and weewx-celestial's nl.conf (the cross-repo test pins it, as for the German and French). Select it with lang = nl in the report.
  • Internal: tests/spec_only_runner.py caught up with 6.5's i18n test renames (it referenced the old German-only test name, so it errored out instead of running).

6.5 2026/07/29

  • A complete French translation ships with the sample LoopData skin (Beta, awaiting native-speaker review -- corrections welcome): lang/fr.conf, including the trend.barometer.desc tendency vocabulary and, for a loopdata instance whose target report is French, the moon phases, body names and all 88 constellation names -- the shared vocabulary in step with weewx-skyfield's and weewx-celestial's fr.conf (a test pins this, as for the German). Select it with lang = fr in the report.

6.4 Release 2026/07/28

ACTION REQUIRED for users who customized [[BarometerTrendDescriptions]]: the [LoopData] [[BarometerTrendDescriptions]] section is REMOVED and now ignored -- delete it from weewx.conf. The trend.barometer.desc descriptions are now gettext-style [Texts] keys in the TARGET report (the English description is the key), so they translate through the same lang-file machinery as everything else -- the shipped lang/de.conf carries all nine in German -- and a custom wording goes on the target report's stanza instead:

[StdReport] [[LoopDataReport]] [[[Texts]]] "Steady" = "Holding"

Users who never touched the section need do nothing: the English defaults are unchanged.

The sample report is translatable, and German ships. The same WeeWX-native machinery as weewx-skyfield 1.12 and weewx-celestial 7.2: every string the page renders is a gettext-style [Texts] key (the English string IS the key, so a missing entry falls back to English one string at a time), lang/en.conf is the reference dictionary, and lang/de.conf is a complete German translation whose shared vocabulary is copied verbatim from those extensions' native-speaker-reviewed files (a test pins the repos together). Everything the live javascript composes -- the LIVE/OFFLINE badge, the dial readout lines, the compass and windrose cardinal letters -- is translated at generation time and fed to the script through json.dumps. Select with lang = de on the report's stanza in weewx.conf.

Two languages meet on the page, and the manual's new Translations page documents the coupling: the page's own labels follow the sample report's lang, but the live VALUES in loop-data.txt (.ordinal_compass, moon_phase, almanac.<body>.label, trend and station fields) render in the language of loopdata's [LoopData] target report. Setting [StdReport] [[Defaults]] lang = de in weewx.conf sets both at once. The shipped lang files carry the target-report sections too (ordinates, moon phases, body and all 88 constellation names, hemispheres), so a German LoopDataReport serves German values.

6.3 Release 2026/07/25

New: station fields. Any $station report tag can be listed as a field with the $ removed: station.uptime.raw, station.uptime.long_form(), station.os_uptime.raw, station.version, station.location, station.altitude.meter.raw, ... Values are evaluated against weewx.station.Station -- the exact object behind the report's $station tag -- with loopdata's usual round(n) and format specs. uptime and os_uptime are recomputed every packet, so a live uptime readout is correct within one packet of a weewxd or server restart (the report-cycle workaround -- extrapolating from an uptime frozen at generation time -- shows the pre-restart uptime still climbing until the next report cycle). Every other $station attribute is constant for the life of the process and is computed once.

A field whose endpoint is a tuple of scalars is now emitted as a json array instead of Python's str() of the tuple: station.latitude renders as ["37", "24.00", "N"] -- the same (degrees, minutes, hemisphere) parts a template indexes -- and an almanac field ending in, e.g., moon_phases gets the same treatment.

6.2 Release 2026/07/24

The sample skin's LIVE indicator now reports a failed loop-data fetch. When the poll for loop_data_file returns anything but the json -- the classic being the web server's 404 page because loopdata writes outside HTML_ROOT (say /dev/shm) with nothing serving it -- the indicator reads "NO DATA (HTTP 404) -- check loop_data_file" (or "BAD DATA -- check loop_data_file" for a 200 that isn't json); a network-level failure shows OFFLINE. Previously the only trace was a JSON.parse error in the browser console and a silently dead panel. A later successful poll rewrites the indicator to LIVE as always. (Same change as weewx-celestial 7.1, which shares the poll's ancestry.)

6.1 Release 2026/07/24

Fixed: a dead or hanging rsync remote could stall loop processing for minutes per attempt despite timeout = 1. That setting only reached rsync's --timeout, which bounds rsync protocol I/O, not the phases ssh owns: a black-holed TCP connect, a stalled handshake, or an auth prompt could each block the LoopProcessor thread far longer (the file kept its full age backlog: written minutes after the packet inside it was stamped, dial pages frozen). LoopData now folds ssh-side bounds into the rsync transport -- ConnectTimeout=<timeout>, ServerAliveInterval=<timeout>, ServerAliveCountMax=2 and BatchMode=yes -- each applied only if your own ssh_options doesn't set that keyword, so anything you set wins. Previously the ConnectTimeout=1 default lived in the ssh_options default value itself, so setting ssh_options at all (a key file, a port) silently discarded it. Setting timeout = 0 disables the time bounds (BatchMode is still applied: an interactive prompt can never be answered under weewxd).

6.0 Release 2026/07/22

Action required if you publish the sample report: upgrading replaces the sample skin's page with the new instrument panel (below) but keeps your existing [LoopData] fields line, and the panel reads fields the old default never included (.raw geometry, unit.label scales, day.windrose.*) -- so with an old fields line the new page serves half-dead: text readouts but no needles, min-max bands or windrose. Replace the fields line under [LoopData] [[Include]] with the one in the README's sample configuration (exactly the fields the panel reads; append any fields other pages of yours use) and restart weewxd. Fresh installs get it by default.

Removed (action required if you used them): the experimental windrun_<dir> observation types (windrun_N ... windrun_NNW) are gone, replaced by the first-class windrose observation below. Fields naming them are now ignored. To migrate: day.windrun_N.sum is element 0 of day.windrose.sum, day.windrun_NNE.sum element 1, and so on clockwise through NNW (element 15). The windrun_<dir> types were always documented as experimental and likely to change; this is that change.

New: windrose. A NOAA-style windrose for any period except current and trend -- including week, month, year, rainyear and alltime, which windrun_<dir> never supported. Per period, loopdata accumulates 16 compass bins x N speed bands, each cell tracking both seconds and distance, plus a directionless calm total (wind below the calm threshold, or no wind direction). Four aggregates project the matrix into json: .sum (16 distances -- windrun parity, unit-convertible, .formatted-able; sub-calm-threshold wind counts as calm time, not distance), .time (16 durations in seconds), .banded (the 16xN seconds matrix -- the NOAA rose), .calm (scalar seconds). Arrays emit json numbers; .round(n) applies per element. A windrose.bands key (band edges, target report windSpeed units) is emitted automatically so page legends never hardcode the bands, and unit.label.windrose gives the distance label. Band edges default to the classic WRPLOT/NOAA bands (0.5, 2.1, 3.6, 5.7, 8.8, 11.1 m/s, shown in report units); override with [LoopData] windrose_bands (ascending edges in the target report's windSpeed unit; the first edge is the calm threshold). Span periods seed at startup with a single SQL aggregate per period, so even alltime.windrose starts fully primed. The sample LoopData skin now draws a live canvas windrose.

Fixed (via the windrose rework): after a weewxd restart, hour-period and continuous-period (1m-1440m/1h-24h) windrun_<dir> buckets started EMPTY and silently filled as packets arrived -- archive-replay priming never carried the bucketed values. windrose accumulators seed every period from the archive at startup.

Changed (check your pages): time-of-event fields (maxtime, mintime, firsttime, lasttime) now format with the field's period as the time context, exactly as WeeWX report tags do -- the target report's [Units][TimeFormats] entry for that period applies. With the standard TimeFormats, day.* times now render as 12:00:00 instead of 07/01/20 12:00:00, hour.* times as 12:00, week.* times as 12:00:00 (Wednesday); month/year/rainyear are unchanged, and alltime uses the year format (the context WeeWX binds for the $alltime tag), also unchanged with the standard TimeFormats. Continuous periods (1m-1440m, 1h-24h) keep the 'current' format. In addition, .formatted on a time-of-event field now yields the same context-formatted time (times carry no label) instead of an unusable raw-number rendering. If your page relied on the old full date-time rendering for day/hour/week times, either adjust the page or set that period's format in the target report's [Units][TimeFormats].

New: unit override in the fields grammar. A field may now name an explicit unit between the aggregation and the format spec (period.obstype[.agg_type][.unit][.format_spec]), exactly as WeeWX report tags allow (e.g., current.windSpeed.beaufort, day.outTemp.avg.degree_C.raw, 10m.windGust.max.knot.raw, trend.barometer.mbar.formatted). Any unit WeeWX knows for the observation's unit group is accepted, regardless of the target report's configured units -- handy for gauges (e.g. Beaufort wind) that expect a fixed unit. The override is on value fields only; the unit.label prefix form is unchanged (WeeWX parity: $unit.label is obstype-only). A unit incompatible with the observation's group causes the field to be omitted.

New: call-syntax format specs. The formatting methods WeeWX report tags can call are now format specs, with the same names, arguments and output: format(format_string, None_string, add_label, localize), nolabel(format_string, None_string), string(None_string) and long_form(format_string, None_string) -- e.g. day.outTemp.maxtime.format("%H:%M"), current.outTemp.format(add_label=False), day.windGust.max.nolabel("%.0f"), day.sunshineDur.sum.long_form(). Arguments must be literals; positional arguments bind exactly as the report tag's do. A bare spec name (day.rain.sum.string) is a zero-argument call, as Cheetah renders it. The unit override composes (day.outTemp.avg.degree_C.nolabel("%.2f")), and the calls work on almanac fields too (almanac.sunrise.format("%H:%M")). string(), or an explicit None_string argument, additionally makes a field with missing data EMIT its None rendering (N/A or the None_string), overriding loopdata's omit-missing-fields default -- handy for pages that want every key present from the first packet. NOTE: a call containing a comma must be quoted in weewx.conf ('day.rain.sum.format("%.2f", add_label=False)') or ConfigObj splits the entry at the comma.

New: round(n). A field may round its value before rendering, exactly as report tags allow -- day.outTemp.max.round(1).raw, or with a unit override, day.barometer.max.mbar.round(1).raw. Ordered between the unit override and the format spec (period.obstype[.agg_type][.unit][.round(n)][.format_spec]); composes with every format spec (round(2).format("%.3f") renders the rounded value) and works on almanac fields (almanac.moon.az.round(1).raw). Most useful with .raw, to publish numbers with limited digits (29.93 instead of 29.927100000000002).

Fixed: the target report's options are now assembled by WeeWX's own build_skin_dict when it exists (loopdata was looking it up under a stale name, so it always fell back to its own assembly), and the fallback now disables ConfigObj interpolation exactly as WeeWX does. Without this, the %(hour)d-style delta-time format strings misparse and long_form() cannot render.

New: the sample LoopData skin is a live instrument panel. A NOAA windrose plus eleven canvas gauges in three rows of four -- temperature, dew point, feels-like and humidity dials wearing today's min-max as a band, a wind compass with a ghost needle at the 10-minute gust direction, a barometer with the 3-hour trend drawn as an arc (chevron showing the direction of travel), rain and rain-rate dials that rescale themselves on a big day, UV, solar radiation and air quality (weewx-purple's pm2_5_aqi) -- in a few hundred lines of dependency-free javascript to crib from, replacing the old tables page. The gauges scale with the window: the engine draws in a 240-unit coordinate system stretched to the css size, so geometry and fonts grow together on a wide display (faces cap at 480px). Gauges for observations the station does not report (UV/radiation/air quality without the sensor, feels-like without appTemp) hide themselves and reappear if the field shows up. The .raw fields drive the geometry, report-formatted fields supply every readout, and unit.label fields pick the dial scales, so the panel follows the target report's units (metric or US) automatically. Fresh installs get the fields line the panel needs; upgrading installs keep their existing fields line -- see the action-required item at the top of this release.

Docs: the README now opens with the point -- report tags updated on every loop packet -- and a four-step example (report tags, the same tags on the fields line, the loop-data.txt they produce, and the javascript that makes the page live).

Internal: the per-aggregate dispatch in add_period_obstype was refactored into per-stats-kind extractor tables (SCALAR_AGGS/VEC_AGGS/FIRSTLAST_AGGS), and the grammar's set of valid aggregate types is now derived from those tables, so the parser and the dispatch can never drift apart. No behavior change.

Internal: the format-spec render block that was repeated in the current, period and trend paths is now a single renderer dispatching through a FORMAT_SPECS table, and the grammar's set of valid format specs (and the almanac grammar's) is derived from that table, just like the aggregate types. No behavior change.

Internal: the test suite moved from bin/user/tests to a top-level tests directory. No change to the installed extension.

5.0 Release 2026/07/19

New: almanac fields. Any WeeWX report almanac tag can now be listed in [LoopData] [[Include]] fields with the $ removed (e.g., almanac.sunrise, almanac.sunrise.raw, almanac.moon_phase, almanac.sun.az, almanac(horizon=-6).sun(use_center=1).rise.raw, almanac.next_solstice.raw). Values are computed with whatever almanac WeeWX has registered (weewx-skyfield, PyEphem, or the built-in fallback) and formatted with the target report's formatter/converter, exactly as the report tag would render. One loopdata extension: almanac(days=±N) evaluates at the same wall-clock time N local calendar days away (e.g., almanac(days=1).sunrise.raw is tomorrow's sunrise). Almanac fields are current-only (no period prefixes or aggregates) and are cached by attribute class: positions/distances recompute every packet, rise/set/transit/visible once per local day, and next_*/previous_* events are kept until the local day advances past the cached event. This makes celestial loop fields available without any packet-injecting service: weewx-celestial's loop fields can be replaced by almanac.* entries.

4.1 Release 2026/07/12

Fix: weewx could fail to shut down if SIGTERM landed while LoopData was starting up or priming its accumulators from the database on the first loop cycle (the broad error handlers there swallowed weewxd's Terminate exception; they now pass it through).

4.0 Release 2026/07/11

No configuration changes are required; this is a drop-in upgrade. The sortedcontainers package is no longer required. Its SortedDict is replaced by MinMaxDict, a minimal min/max-tracking mapping (two heaps with lazy deletion) tailored to the continuous accumulators — faster than SortedDict at every key count, with no pathological worst case. The upstream project has had no release since 2021, and dropping it simplifies installation (no more pip/apt prerequisite step). Continuous accumulators are ~5x faster per packet: the future-debit and per-value timestamp lists are now deques, making the once-per-packet front-removal O(1) instead of an O(window-length) list shift. README: new "Using LoopData in Your Own Skin" section; corrected field examples, config option names and defaults; many typo fixes.

3.9 Release 2026/06/27

Add plumbing for first/last observation types (in case WeeWX fully supports thenm in the future)

3.8 Release 2026/07/27

Many tests added after mutation testing. Support .count which was mistakenly not allowed by the parser

3.6 Release 2026/06/25

Fix three potential latent bugs. Added tests to boost code coverage to 76%. Minus I/O plumbing, test coverage is > 92%.

3.5 Release 2026/06/24

Bug fix for continuous vector accumulator. Code cleanup.

3.4 Release 2026/06/20

Ensure temporary loop data file to actual is atomic.

3.3.2 Release 2022/12/19

Don't try to convert string observations to string.

3.3.1 Release 2022/12/08

Fix for Issue #10. Due to foloating point errors, one could get math domain error for calling sqrt on a negative number. Thanks https://github.com/hoetzgit for reporting this.

3.3 Release 2022/11/23

  1. Rather than just being able to use the 2m, 10m and 24h rolling time period tags, one can now use arbitrary minute tags from 1m to 1440m; and also arbitrary hour tags from 1h to 24h. For example, the maximum windspeed in the last 15 minutes is specified as 15m.wind.max. In the last 17 minutes, it is specified as 17m.wind.max.

3.2 Release 2022/11/22

  1. Lots of code refactoring.
  2. Functionally equivalent to v3.1.

3.1 Release 2022/11/19

  1. Add support for alltime tag.

    Note: The sample skin is updated to use alltime tags. For existing installations that actually use the sample skin, you need to add the following fields to the fields clause in the LoopData section in weewx.conf:

    'alltime.outTemp.max.formatted'
    'alltime.outTemp.min.formatted'
    'alltime.rain.sum.formatted'
    'alltime.wind.rms.formatted'
    'alltime.windGust.max.formatted'
    

3.0.1 Release 2022/10/30

  1. Much more efficient rolling time periods (2m, 10m, 24h, trend).

  2. Add support 24h tag.

    Note: The sample skin is updated to use 24h tags. For existing installations that actually use the sample skin, you need to add the following fields to the fields clause in the LoopData section in weewx.conf:

    '24h.outTemp.max.formatted'
    '24h.outTemp.min.formatted'
    '24h.rain.sum.formatted'
    '24h.wind.rms.formatted'
    '24h.windGust.max.formatted'
    
  3. The python3-sortedcontainers package is now required. To install it on debian, type:

    sudo apt install python3-sortedcontainers
    

2.11 Release 2022/09/25

  1. Specify what drivers this extension has been tested with.
  2. Continue if windrun and/or beaufort can't be calculated.

2.10 Release 2022/02/18

  1. Bugfix. Use loop frequency as the weight when adding to the accumulators.

2.9 Release 13 February 2022

  1. Fix to work with WeeWX 4.6

2.8 Release ?? ??? 2020

  1. Add support hour tag.

    Note: The sample skin is updated to use the hour tags. For existing installations that actually use the sample skin, you need to add the following fields to the fields clause in the LoopData section in weewx.conf:

    'hour.outTemp.max.formatted'
    'hour.outTemp.min.formatted'
    'hour.rain.sum.formatted'
    'hour.wind.rms.formatted'
    'hour.windGust.max.formatted'
    
  2. Add support for 2m tag.

    Note: The sample skin is updated to use the hour tags. For existing installations that actually use the sample skin, you need to add the following fields to the fields clause in the LoopData section in weewx.conf:

    '2m.outTemp.max.formatted'
    '2m.outTemp.min.formatted'
    '2m.rain.sum.formatted'
    '2m.wind.rms.formatted'
    '2m.windGust.max.formatted'
    
  3. In addition to the already supported trend.barometer.desc, trend.barometer.code is also supported. The code can be an integer between -4 and 4, where the value correspsonds to:

    -4 Falling Rapidly
    -3 Falling Quickly
    -2 Falling
    -1 Falling Slowly
     0 Steady
     1 Rising Slowly
     2 Rising
     3 Rising Quickly
     4 Rising Rapidly
    
  4. windrun and beaufort (the deprecated syntax for beaufort) now work correctly.

  5. Experimental feature -- It is likely these observations will be moved to a separate plugin (and available to loopdata). current, 10m, trend, hour and day periods now support windrun_<dir>. That is, the following (which can be used to make a windrose):

    windrun_N
    windrun_NNE
    windrun_NE
    windrun_ENE
    windrun_E
    windrun_ESE
    windrun_SE
    windrun_SSE
    windrun_S
    windrun_SSW
    windrun_SW
    windrun_WSW
    windrun_W
    windrun_WNW
    windrun_NW
    windrun_NNW
    

2.7.2 Release 16 Nov 2020

  1. Fix for customer reported problem. Exception in code when no rainyear fields are included in fields list.

2.7.1 Release 28 Aug 2020

  1. Ignore aggreagate fields base on xtypes as there are no day accumulators from which to reconstruct inital values at start up for these fields.

2.7 Release 11 Aug 2020

  1. Bug fix for accumulators at startup from daily summaries. The daily summary was not being selected in each accumulator range.
  2. Include tests where week_start is 0 (rather than default 6).

2.6 Release 9 Aug 2020

  1. Include sample report LoopDataReport to demonstrate capabilities of the loopdata extension.

2.5 Release 2 Aug 2020

  1. Add support for week, month, year and rainyear.
  2. Add LoopData>LoopFrequency>seconds setting in weewx.conf. This should be set to the number of seconds between loop packets. This is needed to pass the proper weights to the accumulator. This setting defaults to 2.5 (2.5 seconds is the frequency of loop packets for the Davis Vantage Pro 2).
  3. Limit time_delta used for trends to 259200 (3 days).

2.5.b2

Fix for LoopFrequency being ignored (was always using default). Delay accumulator creation until the station time is known (first loop packet).

2.5.b1 Release

  1. Add support for week, month, year and rainyear.
  2. Add LoopData>LoopFrequency>seconds setting in weewx.conf. This should be set to the number of seconds between loop packets. This is needed to pass the proper weights to the accumulator. This setting defaults to 2.5 (2.5 seconds is the frequency of loop packets for the Davis Vantage Pro 2).

2.0.1 Release

  1. Handle case where driver emits packets in a unit_system other than the database unit_system.

2.0 Release

  1. The release is identical to rc0; except for a small type annotation change and additional tests.
  2. Updated documentation (README).

2.0.rc0 07/22/2020

Release Candidate 0

  1. Fix regression in day accumulator introduced in 2.0.b14.
  2. Add test to cover above (added a previous day packet so that tests cover a day boundary.
  3. Only accumulate specified day fields in the day accumulator.

2.0.b15 07/21/2020

  1. Breaking change. If loop_data_dir is a relative path, it is now relative to the target_report directory. After installing, check loop_data_dir before you restart WeeWX.
  2. Default target_report is now WeatherBoard (it was Seasons).
  3. For new installs of WeatherBoard and LoopData, this will now work out of the box for all flavors of WeeWX 4/Py3.7+ installs.

2.0.b14 07/21/2020

[No user facing changes (but please report issues).]

  1. Internal refactoring so that tests use identical code path on each loop packet as production uses.

2.0.b13 07/21/2020

[No user facing changes (but please report issues).]

  1. Clean up dead code.
  2. Simplify creation of target report converter.
  3. Elminate unused archivet period accumulator.

2.0.b12 07/20/2020

  1. Test and fixes for trend.barometer.desc

2.0.b11 07/20/2020

  1. Ability to localize trend.barometer.desc text in weewx.conf.

2.0.b10 07/19/2020

  1. Substantial refactoring. More tests.

2.0.b9 07/19/2020

  1. 10m. is now a full fledged time period (rolling 10 minutes) and supports all that day. supports. For example, 10m.wind.gustdir

2.0.b8 07/18/2020

  1. Save values for trends for only trend fields specified in fields clause.
  2. Only parse fields one time.

2.0.b7 07/18/2020

  1. trend no longer restricted to barometer.

2.0.b6 07/13/2020

  1. Install no longer adds the rename line to weewx.conf. rename is no longer supported. Note: it will not remove rename on existing installations, but rename will be ignored.
  2. Updated README.

2.0.b5 07/12/2020

  1. Fix for case where barometer is reported intermittently.
  2. LoopData now tested in simulator for both intermittent case and when batometer/pressure never reported.

2.0.b4 07/12/2020

  1. If not observation available, the key/value pair will not be present in the json file.

2.0.b3 07/12/2020

  1. Handle stations with no barometer.

2.0.b2 07/12/2020

  1. Watch for None when formatting.

  2. pyflakes and mypy cleanup.

2.0.b1 07/11/2020

  1. Rewritten to use cheetah like identifiers for fields (e.g., $day.rain.sum).

  2. Old names are NOT supported. For existing installations, fields need to be renamed in weewx.conf (LoopData fields), your Cheetah templates/JavasScript.

1.3.19 07/09/2020

  1. Catch and log setup errors. Allow WeeWX to continue with LoopData.

1.3.18 07/09/2020

  1. Handle null barometer values in database.

1.3.17 07/08/2020

  1. Bug fix for wind (not FMT_) version of day stats. The values were not being formatted to the proper number of decimal places.

  2. Add missing highs and lows for vector stats (aka, wind).

1.3.16 07/07/2020

  1. Add missing highs and lows for vector stats (aka, wind).
  2. Add formatting for all times (e.g., FMT_dateTime, FMT_T_HI_wind)

1.3.15 07/06/2020

Fix OutOfSpan error caused by creating accumulator with current time rather than packet time.

1.3.14 07/06/2020

  1. Add support for VectorStats: Wind Avg, Wind RMS, Wind Vector Avg, Wind Vector Dir.
  2. More tests written.

1.3.13 07/05/2020

Fall back to using copy.deepcopy for earlier WeeWX 4 versions.

1.3.12 07/05/2020

  1. Properly import weeutil.config

1.3.11 07/05/2020

  1. pyflakes found some issues including lots of unused imports.
  2. Update one test.

1.3.10 07/05/2020

  1. Redo how 10m max gust is handled when windGust not in packet. Simply substitue windSpeed for windGust.
  2. Test overriding km_per_hour label in weewx.conf Defaults>Units.
  3. Add test of CC3000 packets.
  4. Add test of IP100 packets.
  5. Add test of Simulator packets.

1.3.9 07/05/2020

  1. Properly construct dictionary of target_report so formatting and conversion actually does what it should.
  2. Fix a problem reported by a user running pyflakes. USUnit wasn't qualified.
  3. Fix an issue reported. windGust was assumed to be in the packet. Now, if not present, max windSpeed in the archive period accumulator stats is used to compute 10m gust.

1.3.8 07/05/2020

Require at least Python 3.7. Handle None for windSpeed and and windDir in the database (as some drivers write None rather than omitting). More tests.

1.3.7 07/04/2020

Refactor loopdata.py to facilitate testing.

1.3.6 07/04/2020

Add tests for conversions and formatting: weewx.METRIC -> metric reports weewx.METRIC -> us reports weewx.US -> metric reports weewx.US -> us reports weewx.METRICWX -> metric reports weewx.METRICWX -> us reports

1.3.5 07/03/2020

Converting to report units was being done correctly for observations. This release fixes conversions for highs, lows, sums, averages and weighed averages.

1.3.4 07/02/2020

Installer uses FMT_SUM_rain rather than FMT_day_rain_total in LoopData section of weewx.conf. day_rain_total is not emitted by many drivers and was a source of confusion.

1.3.3 07/02/2020

Refrain from trying to format the sum of unix_epoch observations.

1.3.2 07/02/2020

Catch FMT_SUM_ formatting exceptions and print error to log with obstype, unit_type and unit_group.

1.3.1 06/30/2020

Fixed issue when specifying port in rsync section. RsyncUpload was expecting a string for port, loopdata was passing an int.

1.3 06/29/2020

Introduced AVG_/FMT_AVG_ and WAVG_/FMT_WAVG_ prefixes for today's average and weighted average of an observation, respectively.

1.2.2 06/28/2020

Introduced SUM_ and FMT_SUM_ prefixes for today's totals of observations. This is most useful for rain, where SUM_rain (or FMT_SUM_rain) will yield today's cumultive rainfall.

1.2.1 06/27/2020

Inexplicably, just a version bump. No other changes.

1.2 06/27/2020

Fixed issue where loop thread was created multiple times.

1.1 06/27/2020

Fixes to work with Metric/Mixed database and to correctly target non-US reports.

1.0 01/19/2020

Initial release of loopdata.