feat: add ability to downgrade aur package using git #260

darukutsu · 2025-11-21T14:48:41Z

Checklist

Admin:
- No duplicate PRs
Integration:
- Always add unit tests for fixes or new functionality
- If adding new options, update dist/completion/*
Documentation:
- If adding new options, update usage and doc/*.ronn
- If adding new strings, update locale/*.po

Description

If you don't store all tarballs in your ~/.cache/yay (like me, store only last 2) or if you use other aur-manager you could instead use vcs to switch to desired branch and rebuild chosen version. Hopefully everything works as should, tested only for yay.

src/downgrade

darukutsu · 2025-11-22T12:24:28Z

Added some tests, however i don't understand yet how syntax works, so will edit them when I figure out.

src/downgrade

test/search_git/specific-version.t

pbrisbin

This is looking good. Sorry, I should've warned you not to worry about the existing shfmt / shellcheck stuff.

src/downgrade

darukutsu · 2025-12-30T14:11:57Z

I realized that makepkg -fs outputs correct pkg name so no more need to rename and rename back to use gitpkg... now its literally placeholder for build_pkg() to determine whether it should be build. File with gitpkg never exists. So only what we need to alter is to_install array and I come up with safe solution. That previous renaming using -ctime wasn't imo very good solution.

see

593: printf "%s\0" "$item" >> new_to_install.downgrade

613: mapfile -d '' to_install < new_to_install.downgrade
614: rm new_to_install.downgrade

Here's simple command how to try build something from aur

 $ downgrade --git --pacman-cache ~/.cache/yay --maxdepth 2 'ani-cli>4.7' bash

pbrisbin

Left some thoughts, but nothing blocking. Let me know if you'd like to address them before merging.

src/downgrade

test/search_git/specific-version.t

test/helper.sh

doc/downgrade.8.ronn

pbrisbin · 2026-01-05T15:51:17Z

I'm going to push a commit that improves the "No fixture found" error message, and then does the thing it suggests to fix those. The remaining failures seem like they may just be actual bugs.

test/bin/act-like

pbrisbin · 2026-01-05T17:17:11Z

Looking at the test failures, I see a few examples like:

   > search_ala 'xorg-twm' | filter_packages 'xorg-twm' '=' '1.0.8-1'
   https://archive.archlinux.org/packages/x/xorg-twm/xorg-twm-1.0.8-1-x86_64.pkg.tar.xz
+  https://archive.archlinux.org/packages/x/xorg-twm/xorg-twm-1.0.8-2-x86_64.pkg.tar.xz
+  https://archive.archlinux.org/packages/x/xorg-twm/xorg-twm-1.0.8-3-x86_64.pkg.tar.xz

This test is really about = and ensuring that the results are exactly and only 1.0.8-1. The + lines indicate this has been broken since the 1.0.8-2 and -3 lines are being returned unexpectedly.

Is it possible your regex changes are causing the operators filtering to no longer work?

darukutsu · 2026-01-05T17:40:50Z

I fixed regex already, however I don't understand why search_git is not outputting anything. It should, since we clone repo and set correct path to CacheDir. Once I figured that out I push fixes.

darukutsu · 2026-01-05T18:01:08Z

I'm idiot I forgot to read config values now it should be without error.

darukutsu · 2026-01-08T09:58:38Z

hopefully i rebased correctly, first time doing this :D

darukutsu · 2026-01-08T10:01:05Z

dont run workflows yet, I just discovered just test from your readme will fix issue then after next commit

darukutsu · 2026-01-08T10:39:54Z

So this part should match multiple lines so when you update we don't have to rewrite test to add new version however it matches only first match. So i assume cram does not support multiline regex match. Any suggestion?

  $ search_git 'downgrade' | filter_packages 'downgrade' '>' '6.0.0'
  /tmp/.*/downgrade-(?!6\.0\.0)[0-9.]+-.*-any.gitpkg.tar.gz (re)
  /tmp/.*/downgrade-([6-9]|[1-9][0-9]+).[0-9.-]+-.*-any.gitpkg.tar.gz (re)

darukutsu · 2026-01-08T11:34:47Z

Ok so come up with workaround to use grep -vP and if there something that isn't matched it will spew out as unmatched line. Or if don't like this solution maybe switch to bats as suggested in #170 but I have no idea if they have multiline support. There is still issue with su needs authentication. Maybe fake su binary that does nothing. It depends whether container is root by default or not. If it is, we have switch user so we can actually use makepkg otherwise just hide su binary.

pbrisbin · 2026-01-08T14:01:10Z

So i assume cram does not support multiline regex match. Any suggestion?

That's correct. Cram is basically golden testing, so it should be written to match all the expected output. The glob and regex form can be useful, but should be used sparingly.

I would probably use tail (or head depending on how it sorts) to filter to the oldest 3 (or so) tags, then you can just match exactly because that output should remain stable even when new tags are added.

  $ search_git 'downgrade' | tail -n 3
  /tmp/*/downgrade-3.2-1-any.gitpkg.tar.gz (glob)
  /tmp/*/downgrade-3.1-1-any.gitpkg.tar.gz (glob)
  /tmp/*/downgrade-2.1-1-any.gitpkg.tar.gz (glob)

Does that work?

There is still issue with su needs authentication. Maybe fake su binary that does nothing

I'm personally OK if we just don't test that function, it seems hard to get any useful test coverage there. It's an add-on / special case thing that, if it breaks, downgrade can still be used generally.

darukutsu · 2026-01-08T14:28:36Z

did i mess up something?

darukutsu · 2026-01-08T14:35:36Z

just test passing on my side don't understand why did this happen

pbrisbin · 2026-01-08T14:44:33Z

OK, I see what's happening. If you are happy with it now, I can take it from here and push some fixes before merging.

On GHA, tests that run in containers must use `root`[^1]. Otherwise, you don't have access to write in the workspace. For example, our tests failing couldn't write the `.t.err` files. Therefore, to use the `archlinux` container, we have to remove this test that was asserting errors when running as non-`root`. In testing that this would work, I used `sudo just test` locally, which ran into an unrelated error being unable to write in `/tmp`. This is probably solvable, but we never used that file, so I replaced it with `/dev/null`. Lastly, I installed `git` since I know we'll need that. [^1]: https://docs.github.com/en/actions/reference/workflows-and-actions/dockerfile-support#user

This is fully opt-in by the `--git` option (or equivalent env var). If enabled, we check to see if the configured cache directory represents the git repository. This would be the case if the user is downgrading an AUR package installed via `yay`. In such a repository, we can hunt for commits that represent the older versions of the package and present them as options to downgrade to. If selected, we can check out that commit and build the package at that version (if necessary) and install it. The logic is isolated to a `search_git` function, for handling locating commits that represent versions and output potential package names that include "gitpkg" in them, and a `build_pkg` function that looks for such package names and builds them. We have to be careful to use `su` so that `makepkg` is not run as root. The only changes to `downgrade` outside of these new functions is to regexs used to locate and parse package names, but this is well covered by tests.

pbrisbin · 2026-01-08T16:02:50Z

OK, I'm done and ready to merge. I will let this sit for a few hours in case you want to review my changes, commit messages, etc.

Thanks for the contribution, I hope it works well for you!

darukutsu · 2026-01-08T16:15:38Z

lgtm

darukutsu · 2026-01-08T17:18:39Z

there is issue restyle fucked up something probably, quickfix:

    pairs=$(git -C "$path" log -p --oneline --format="GIT: %h" PKGBUILD | grep -e 'GIT:.*' -e '+pkgver' | grep -Pzo 'GIT: .*\n\+pkgver=.*\n' | sed 's/GIT: \|+pkgver=//' | tr -d '\0')

darukutsu · 2026-01-08T17:20:11Z

for some reason missing quoted didn't saw it in head diff when was merging

darukutsu · 2026-01-08T17:21:59Z

how did tests pass tho?

pbrisbin · 2026-01-08T17:23:24Z

restyle fucked up something probably

This was what Restyled did, it was only whitespace.

diff --git a/src/downgrade b/src/downgrade
index 84e9bc1..24a3204 100755
--- a/src/downgrade
+++ b/src/downgrade
@@ -211,8 +211,8 @@ search_git() {
     path="$cache/$name" # directory name
     if ! [[ -d "$path/.git" ]]; then
       {
-	gettext "\`$path\" not a git directory, skipping..."
-	echo
+        gettext "\`$path\" not a git directory, skipping..."
+        echo
       } >&2
       continue
     fi
@@ -586,12 +586,11 @@ build_pkg() {
       exit 1
     fi
 
-
     item="$(find "$path" -maxdepth 1 -type f -name "$(sed -nr '/.*Finished making: ([^ ]+) ([^ ]+) .*/{s//\1-\2/;p}' makepkg.log)*" -print -quit)"
     rm makepkg.log
     popd >/dev/null
   fi
-  printf "%s\0" "$item" >> new_to_install.downgrade
+  printf "%s\0" "$item" >>new_to_install.downgrade
 }
 
 cli() {
@@ -611,7 +610,7 @@ cli() {
     for item in "${to_install[@]}"; do
       build_pkg "$item"
     done
-    mapfile -d '' to_install < new_to_install.downgrade
+    mapfile -d '' to_install <new_to_install.downgrade
     rm new_to_install.downgrade
   fi
 
-- 
2.52.0

how did tests pass tho?

🤷

darukutsu · 2026-01-08T17:58:19Z

5f75463 this was commit that changed something in downgrade (and is correct) so maybe build?
Edit: yea so everything in repo is correct build must be broken

darukutsu · 2026-01-08T18:02:32Z

to be clear:

                                                                                  v --- `missing quote`
    pairs=$(git -C "$path" log -p --oneline --format="GIT: %h" PKGBUILD | grep -e GIT:.*' -e '+pkgver' | grep -Pzo 'GIT: .*\n\+pkgver=.*\n' | sed 's/GIT: \|+pkgver=//' | tr -d '\0')

pbrisbin · 2026-01-08T18:21:25Z

build must be broken

You are right. The m4 process that injects version must be mangling quotes. I feel like I've solved that before.

I can try to update the tests so it builds before testing to catch these.

pbrisbin · 2026-01-08T18:31:38Z

I'm going to revert and release since this breaks everything. Will bring it back after addressing the CI gap.

pbrisbin · 2026-01-08T21:22:56Z

FYI the feature was reverted in 11.7 and brought back in 11.8.

pbrisbin reviewed Nov 21, 2025

View reviewed changes

src/downgrade Outdated Show resolved Hide resolved

src/downgrade Outdated Show resolved Hide resolved

pbrisbin reviewed Nov 21, 2025

View reviewed changes

src/downgrade Outdated Show resolved Hide resolved

pbrisbin reviewed Nov 24, 2025

View reviewed changes

pbrisbin reviewed Nov 25, 2025

View reviewed changes

src/downgrade Outdated Show resolved Hide resolved

pbrisbin approved these changes Jan 5, 2026

View reviewed changes

pbrisbin reviewed Jan 5, 2026

View reviewed changes

test/bin/act-like Show resolved Hide resolved

darukutsu force-pushed the aur-git branch from 7dd8cf9 to 518fa1c Compare January 8, 2026 09:56

darukutsu force-pushed the aur-git branch from 162affe to 5d3f042 Compare January 8, 2026 14:32

pbrisbin and others added 2 commits January 8, 2026 10:51

pbrisbin force-pushed the aur-git branch from 01f99c7 to 5f75463 Compare January 8, 2026 16:00

pbrisbin merged commit 3fd2cdc into archlinux-downgrade:main Jan 8, 2026
3 checks passed

feat: add ability to downgrade aur package using git #260

feat: add ability to downgrade aur package using git #260

Uh oh!

Conversation

darukutsu commented Nov 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Checklist

Description

Uh oh!

Uh oh!

Uh oh!

Uh oh!

darukutsu commented Nov 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

pbrisbin left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

darukutsu commented Dec 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pbrisbin left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

pbrisbin commented Jan 5, 2026

Uh oh!

Uh oh!

pbrisbin commented Jan 5, 2026

Uh oh!

darukutsu commented Jan 5, 2026

Uh oh!

darukutsu commented Jan 5, 2026

Uh oh!

darukutsu commented Jan 8, 2026

Uh oh!

darukutsu commented Jan 8, 2026

Uh oh!

darukutsu commented Jan 8, 2026

Uh oh!

darukutsu commented Jan 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pbrisbin commented Jan 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

darukutsu commented Jan 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

darukutsu commented Jan 8, 2026

Uh oh!

pbrisbin commented Jan 8, 2026

Uh oh!

pbrisbin commented Jan 8, 2026

Uh oh!

darukutsu commented Jan 8, 2026

Uh oh!

Uh oh!

darukutsu commented Jan 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

darukutsu commented Jan 8, 2026

Uh oh!

darukutsu commented Jan 8, 2026

Uh oh!

pbrisbin commented Jan 8, 2026

darukutsu commented Nov 21, 2025 •

edited

Loading

darukutsu commented Nov 22, 2025 •

edited

Loading

darukutsu commented Dec 30, 2025 •

edited

Loading

darukutsu commented Jan 8, 2026 •

edited

Loading

pbrisbin commented Jan 8, 2026 •

edited

Loading

darukutsu commented Jan 8, 2026 •

edited

Loading

darukutsu commented Jan 8, 2026 •

edited

Loading

darukutsu commented Jan 8, 2026 •

edited

Loading

darukutsu commented Jan 8, 2026 •

edited

Loading