Skip to content

Commit d65c48c

Browse files
committed
docs(s002): correct the cause named in the previous commit
699a9b1 said gh run download refuses the two-path artifact layout with "would result in path traversal", measured against a pre-slice-002 artifact that downloaded cleanly. The layout was not the cause and the measurement was not a comparison: the old artifact had been fetched with -D and the new one without it. Run the same way, gh 2.45.0 raises that error for either artifact when it extracts into the current directory, and succeeds for either with -D; gh 2.63.2 succeeds for both with or without. It is a bug in the client version Ubuntu ships, it predates this slice, and nothing here caused or fixed it. The layout change stands, on the reason that survives: both files at the root of the artifact is the shape every earlier release used, and moving the binary under burrito_out/ would make consumers absorb a change for no benefit. The comment in package.yml is corrected in place rather than deleted, because a record that quietly loses a wrong claim teaches nothing. Signed-off-by: Ayla Croft <aylacroft@proton.me>
1 parent 699a9b1 commit d65c48c

1 file changed

Lines changed: 12 additions & 6 deletions

File tree

‎.github/workflows/package.yml‎

Lines changed: 12 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -366,12 +366,18 @@ jobs:
366366
path: burrito_out
367367

368368
# The bill is placed beside the binary rather than listed as a second upload path, so that
369-
# both land at the root of the artifact. Uploading the two paths separately made
370-
# upload-artifact take the workspace as their common ancestor and file the binary under
371-
# `burrito_out/` inside the zip, and `gh run download --name` on the gh in Ubuntu's archive
372-
# (2.45.0) then refuses the whole artifact with "would result in path traversal". Measured
373-
# on run 35898448972 against a pre-slice-002 artifact that downloads cleanly with the same
374-
# client. Whoever is at fault, a reviewer's first attempt to fetch a binary must not fail.
369+
# both land at the root of the artifact. Listing them separately makes upload-artifact take
370+
# the workspace as their common ancestor and file the binary under `burrito_out/` inside the
371+
# zip, which is a layout change every earlier release's consumers would have to absorb for
372+
# no benefit.
373+
#
374+
# Corrected 2026-09-23, because the first version of this comment named a cause that is not
375+
# true. It said `gh run download --name` refuses the two-path layout with "would result in
376+
# path traversal". It does refuse, but the layout has nothing to do with it: gh 2.45.0 (the
377+
# version in Ubuntu's archive) raises that error for any artifact when extracting into the
378+
# current directory, including a pre-slice-002 single-file one, and passing `-D <dir>`
379+
# succeeds for both. gh 2.63.2 downloads both without `-D`. The error was measured against
380+
# unlike invocations and read as a regression it was not.
375381
- uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
376382
with:
377383
name: ${{ matrix.artifact }}

0 commit comments

Comments
 (0)