Skip to content

Introduce sms-api & qt-sms - #442

Draft
OrkunTokdemir wants to merge 177 commits into
qt-labs:devfrom
OrkunTokdemir:orkun_sms_api_12_03_2026
Draft

OrkunTokdemir wants to merge 177 commits into
qt-labs:devfrom
OrkunTokdemir:orkun_sms_api_12_03_2026

Conversation

@OrkunTokdemir

Copy link
Copy Markdown
Collaborator

No description provided.

@OrkunTokdemir
OrkunTokdemir force-pushed the orkun_sms_api_12_03_2026 branch from 090c53f to 4d27dd8 Compare March 16, 2026 16:17
@OrkunTokdemir
OrkunTokdemir force-pushed the orkun_sms_api_12_03_2026 branch 3 times, most recently from 41d9cd4 to 1e8b129 Compare April 8, 2026 13:29
@OrkunTokdemir
OrkunTokdemir force-pushed the orkun_sms_api_12_03_2026 branch from 9eb50c7 to 3390a3a Compare April 10, 2026 17:01
@OrkunTokdemir OrkunTokdemir changed the title Introduce sms-api Introduce sms-api & qt-sms Apr 13, 2026
@OrkunTokdemir
OrkunTokdemir force-pushed the orkun_sms_api_12_03_2026 branch 3 times, most recently from 45ca481 to 576afe3 Compare April 22, 2026 08:20
@OrkunTokdemir
OrkunTokdemir force-pushed the orkun_sms_api_12_03_2026 branch 2 times, most recently from b15546d to 51b0acc Compare May 8, 2026 14:47
@OrkunTokdemir
OrkunTokdemir force-pushed the orkun_sms_api_12_03_2026 branch 3 times, most recently from d895bee to f0f8012 Compare June 1, 2026 11:26
@OrkunTokdemir
OrkunTokdemir force-pushed the orkun_sms_api_12_03_2026 branch 2 times, most recently from 0ad368d to 7c07a83 Compare June 5, 2026 15:47
@OrkunTokdemir
OrkunTokdemir force-pushed the orkun_sms_api_12_03_2026 branch 7 times, most recently from e6706ad to 1cf6fe4 Compare June 24, 2026 14:53
@OrkunTokdemir
OrkunTokdemir force-pushed the orkun_sms_api_12_03_2026 branch 3 times, most recently from a1f37e2 to 21bc761 Compare June 25, 2026 14:49
@OrkunTokdemir
OrkunTokdemir marked this pull request as ready for review July 17, 2026 11:39
@OrkunTokdemir
OrkunTokdemir marked this pull request as draft July 17, 2026 11:39
@OrkunTokdemir
OrkunTokdemir force-pushed the orkun_sms_api_12_03_2026 branch 2 times, most recently from ee6e51e to b253aba Compare July 21, 2026 08:36
On Linux and macOS the service puts qtclient_socket into the temporary
directory, but a Unix socket path holds at most 103 bytes. When the
temporary directory is too long for that, for example in a sandbox, the
service now listens in XDG_RUNTIME_DIR instead, or in /tmp with the user
ID appended when that is not set or is also too long. The client resolves
the socket the same way, so the two still meet.
The service no longer keeps a current installation directory between
requests. Each request that works on an installation names it, with
installRoot or with a workingDirectory inside a recorded installation,
and fails with error 9001 when it names none. Packages.install,
fetchRequirements, searchAvailablePackages and listInstalledPackages
take the installation target first, like the C++ client API. A search
that names an installation reports the packages installed in it as
installed, so callers no longer need a separate list to tell them apart.

listInstalledPackages resolves the installation root next to the
packages and takes the packageType filter, which lists the alias
packages alone. New requests: packages/installations lists the recorded
installations, and packages/validate-root checks a directory before an
install without storing it. Packages carry packageType and
installReason, and the error codes 3001, 7001 and 9001 are known. The
installationPath setting is stored only: its helpers stay, documented
as such.

The integration test installs into its own directory explicitly and
lists it afterwards.
The service implements packages/remove and packages/purge now.
Packages.remove removes the named packages from the installation the
target names; the packages that depend on them go too, and autoRemove
also takes the dependencies that nothing needs any more. removeOrphans
sends the same request without packages, which removes only those.
purge removes a complete installation and resolves the installations
that are left.

When a removal takes packages the client did not name, the service asks
the removal_confirmation question through service/question, listing
them with the reason, unless confirmRemoval says the user has confirmed
already. The question id is a constant next to the method names so
clients can show it as a confirmation rather than a plain choice. The
removal options are sent only when they are true, like the C++ client
sends them, and error 8002 is known.
The service keeps no installation root between requests any more, so
that applications sharing it, such as Qt Creator and Clate, cannot
work on each other's installation by accident. Instead of pushing the
configured installationPath into the service on every connect, the
extension now names the root in each request: the search, the
requirements check and the install all carry it, and the newest
framework check does too.

A search that names the root reports the packages installed there, so
the separate packages/list requests that built the installed set are
gone. "Set installation path" checks the directory with
packages/validate-root, which stores nothing, before saving it, and a
path changed in the settings is checked the same way and reported when
the service cannot install into it. The install offers to set the path
when the setting is empty. The folder dialog opens at the resolved
root, not at the raw ~ setting.
The trash icon of the Installed Packages view deleted the version
folder from disk, because the service could not remove packages yet.
It now sends packages/remove for the alias packages of the version,
read from packages/list with the packageType filter, with autoRemove so
that the dependencies nothing needs any more, such as the build tools,
go too. The service asks the removal_confirmation question when the
removal reaches beyond the named packages; the extension shows it as a
modal dialog that lists them with the reason, and dismissing it answers
the cancel choice. Progress and the messages of the service are shown,
a declined confirmation stays quiet, and the tool paths are re-detected
afterwards because CMake or Ninja may have gone with the version.

A version the service does not record, for example one installed by
another tool, cannot be removed through it. Its folder is still deleted
from disk, after a dialog that says so.

The install-in-progress flag is now a package-operation flag, set by the
removal too: the service runs one package operation at a time, and the
disk watchers ignore the churn of a removal like that of an install.
Backend requests fail with error 3000 when the service has no valid
credentials or cannot renew them, and with 3001 when the backend
rejects them; before, these surfaced as a network error. A search or
install that fails this way now shows what happened and offers to sign
in, because signing in again is what helps, instead of a plain service
error.
The Installed Packages view read the version folders from disk, because
the service kept its own idea of the installation root. Now that every
request names its root, the view asks the service: packages/list with
the packageType filter gives the alias packages installed in the
configured root, grouped by Qt version and shown with their packages,
such as the essentials and the add-ons.

The service is started right after the activation, without blocking
it, and the view shows a progress bar until the service is connected
and has answered; while the bootstrap still installs or starts the
service, the view says so. A failure, for example a journal written by
an older service, is shown in the view with the message and a Retry
item instead of a notification, and a refresh action sits in the view
title. The view is always visible now, with welcome text when nothing
is installed, because hiding it would hide the progress too.

Installs and removals made outside the extension are picked up through
the disk watcher, which re-reads the service once the events settle.
The folder deletion fallback of the removal is gone: the view lists only
versions the service records.
…view

While the first Qt version downloaded, the view said that no Qt
versions are installed: the disk watcher stays quiet during a package
operation and the service lists the version only once it is done. The
view now shows the running install or removal as an item with a
spinner at the top, together with its progress bar, until the operation
settles and the view reads the service again.
The service asks its removal_confirmation question when a removal
takes packages the client did not name, and with autoRemove that is
every module of the version. The extension showed that list in a modal
dialog, which for a Qt version runs to hundreds of lines and blocks
the window.

The extension already asks before it sends the removal, and that
dialog says that the packages nothing else needs go too, so the
request now carries confirmRemoval and the service does not ask again;
the log lists what was removed. The question handler stays for a
service that asks anyway: it shows the count and the first entries and
writes the full list to the log.
The autoRemoveUnneeded setting, on by default, decides whether a
removal also takes the packages that nothing else needs any more, such
as the modules of the version and the build tools it brought along.
The confirmation dialog is one short line: it asks about the version
and says that unneeded packages go too when the setting is on. The
fallback for a service that asks its own confirmation shows only the
count and leaves the list to the log.
The removal dialog said that the packages nothing else needs go too.
That is the autoRemoveUnneeded setting, not something to read at every
removal, so the dialog asks only "Remove Qt X?" and the log says what
the removal took.
The usage data and crash dump checkboxes of the walkthrough's Privacy
& data card were checked until the user unchecked them. They now start
unchecked, also after a walkthrough reset, so nothing is collected
unless the user opts in. The usage data choice is forwarded to the
service as its telemetry setting as before, so the service gets
"false" until the box is checked.
The view listed Qt versions without saying where they are installed.
The root now sits next to the view title, with the home directory
shortened to ~: the configured root while the view loads, and the root
the service resolved once it has answered. It follows a change of the
installation path, because that refreshes the view.
The license panel focuses the accept checkbox, and Enter ticks it, but
the focus stayed on the checkbox, so Continue needed the mouse or the
arrow keys. Ticking the checkbox now moves the focus to Continue once
it is enabled, and the focused Continue is outlined also after a mouse
click, so pressing Enter twice accepts the agreements.
…front

The notification after an install offered "Start Walkthrough" also
when the install was started from the walkthrough and the walkthrough
was still the active editor. The button and its sentence are now left
out in that case; "Open Qt Panel" stays.
When we don't pass that argument, the bootstrap installer will fall bac
to the old usage which is `~/QtAlpha`
The operation item at the top of the view only had a spinner, so the
download and install progress was visible in the notification alone.
The item now says whether the package is downloading or installing and
shows a text progress bar with the percentage. Only the item re-renders
on a progress update, and only when what it shows changes.
* Only show the Qt version without its packages
* Add package icon to the Qt version tree item
…ages

The installation root was shown next to the view title, shortened with
a tilde, where it was hard to read. It is now the first item of the
list with its full path, in every state of the view. An inline button
on the item reveals the root in the file explorer of the system. When
no installation path is set, the item offers to set it.

Since the list is never empty any more, the welcome content of the
view is replaced by items. The view shows a spinner item while it
reads the installed versions, and says that no Qt version is installed
with an Install Qt item when the list is empty.
@OrkunTokdemir
OrkunTokdemir force-pushed the orkun_sms_api_12_03_2026 branch from 76a28f2 to 6e6c8fa Compare November 18, 2026 13:53
Replace the individual exports (Home, IsWindows, IsMacOS, IsLinux,
IsUnix, IsArm64, IsArm32, Isx86, Isx64, OSExeSuffix, UserLocalDir)
with a single host object.

Call sites now only need to import `host` instead of each constant separately,
which simplifies the import statements.

Also rename `Home` to `userHome` to be consistent with `userLocalDir`.

Amends 1fe57a1
The service now installs every product into a folder per version below
the folder of its vendor, for example io.qt/qtframework/6.12.0,
org.cmake/cmake/3.30.6 and org.ninja-build/ninja/1.12.1. The extension
still looked for QtFramework/<version> and Tools/, so after an install
the framework step of the walkthrough never completed and the last
step stayed locked. For the same reason the installed Qt was not
registered as a kit, and CMake and Ninja were not published to qt-core
and qt-cpp.

The framework step now follows the Qt versions the service lists with
packages/list. The Installed Packages view shares each answer it reads,
and the walkthrough re-renders when the listed versions change, so the
step no longer depends on the folders on disk.

The paths that have to be found on disk, qtpaths, CMake, Ninja and the
folders to watch, are built in install-layout.ts, the one place that
knows the layout.
Removing the only Qt version made the service forget the installation
root, while files that belong to no package stayed in the folder. The
service refuses to install into a folder that is not empty and that it
does not record, so the next install failed.

When the removal would leave no package in the root, packages/purge is
now sent in place of packages/remove, which removes the folder as well.
The confirmation dialog says that the folder is deleted in that case.
The Create Account buttons opened login.qt.io/register, the old
registration page, while Sign In already showed the new Qt Account
page of the login server. That page has a Create Account tab next to
Sign In, and the login server has no address that opens on the tab.

Create Account now starts the browser sign-in as Sign In does. The
new account is signed in when the browser comes back, so the user is
no longer left in the browser after the email validation.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Development

Successfully merging this pull request may close these issues.

1 participant