Re: [PATCH] [pre-commit] Add shellcheck

[Tom de Vries <tdevries@suse.de>]

On 6/19/26 7:56 PM, Simon Marchi wrote:
> 
> 
> On 2026-06-19 07:19, Tom de Vries wrote:
>> On 6/18/26 7:05 PM, Simon Marchi wrote:
>>>
>>>
>>> On 2026-06-18 11:19, Tom de Vries wrote:
>>>> I found a pure python implementation of shellcheck [1].
>>>>
>>>> Use it to run shellcheck on scripts in the repo.
>>>>
>>>> Exclude any scripts that are not currently clean.
>>>>
>>>> Running it seems reasonably fast:
>>>> ...
>>>> $ pre-commit run shellcheck --all-files -v
>>>> shellcheck...............................................................Passed
>>>> - hook id: shellcheck
>>>> - duration: 0.06s
>>>> ...
>>>>
>>>> For information on other solutions, see this RFC [2].
>>>>
>>>> [1] https://pypi.org/project/pureshellcheck/0.2.2/
>>>> [2] https://sourceware.org/pipermail/gdb-patches/2024-November/213400.html
>>>
>>> While I'm sympathetic to the use of pre-commit (I added the first
>>> hooks), I'm starting to get a bit worried that we kind of blindly pull
>>> hooks from random places without paying much attention.  This is going
>>> to get executed on the machines of many GDB devs, and probably some CI
>>> too.
>>>
>>> I had this thought because this one has the "random project on github
>>> vibes" (it appears to be a vibe coded project started a week ago).  More
>>> established projects (like black) are not immune to being compromised,
>>> but they are easier to trust I guess.
>>>
>>> Assuming you gave a quick look at the code of this project and judge
>>> that it's fine,
>>
>> Yeah, I did, that is, I cloned the github repository, checked it out at main/v0.2.2, and asked claude code to review it on safety aspects.
> 
> Cool, thanks.
> 
>>> how can we ensure that whatever pre-commit pulls is what
>>> you reviewed?
>>
>> The mechanism to get the tool proposed in this patch is via pip.  The pypi index mentions the github repo as source, so I'm relying on that.
> 
> Ok I didn't get this at first.  You specify the tool (and exact version)
> using additional_dependencies, that downloads it in the venv, and then
> it runs "pureshellcheck" since it's specified as the entry point.
> 
> I previously thought it would get it from the github repo directly, but
> no (the github repo URL isn't even mentioned anywhere in the hook).
> 
> Getting a specific version from pypi seems safe-ish, in that it's not
> possible (from what I've read) to swap a file with another, keeping the
> same name.  Although I read that it might be possible for someone to
> upload a source package at first, and then only later upload a binary
> one for the same version (which could differ and contain something
> nasty).  pip would then prefer the binary one as soon as it's uploaded.
> 
>  From what I understand, with hooks like flake8, pre-commit clones the
> specified ref, then does "pip install" in it to install it in the venv.
> So we use the flake8 code as specified by that ref, but its dependencies
> are still downloaded from pypi, we can't get around that.
> 
>>> We use a git tag, but is that sufficient?  Could a
>>> (malicious or compromised) project publish a tag, and then replaced that
>>> tag with something else later?
>>
>> I suppose it's possible.
> 
> I found out about
> 
>    $ pre-commit autoupdate --freeze
> 
> it transforms:
> 
>      rev: 26.5.1
> 
> into
> 
>      rev: 4160603246a6b365d4a2af661c6d71b0a0f50478  # frozen: 26.5.1
> 
> It won't solve all the integrity concerns, but we might as well use
> that, I don't see a downside.
>

Ack, that's what I proposed to use in v2.

Note that the --freeze only works until the next autoupdate without 
freeze.  So we'll likely need an autoupdate script that skips 
pureshellcheck, along the lines of:
...
$ cat contrib/pre-commit-autoupdate.sh
#!/bin/bash

repos=($(grep "\- repo: " .pre-commit-config.yaml \
	     | awk '{print $3}'))

for repo in "${repos[@]}"; do
     if [ "$repo" = local ]; then
	continue
     fi

     pre-commit autoupdate \
	       --repo "$repo"
done
...

>>> An alternative could be to point to a specific commit hash.  A more
>>> radical alternative would be to vendor (put in our repo) the code of the
>>> hooks we use.
>>>
>>
>> Yeah, that is safer.  It would mean though using github as the source. But the repo as is doesn't have pre-commit hooks.  I worked around that before by using my own github account (see commit 7f6c7a5bb37 ("[pre-commit] Add tclint hook")).  I'll submit a v2 shortly that uses this approach.
>>
>> FWIW, I'm open to any other solutions.
> 
> Could you maybe submit a .pre-commit-hooks.yaml to that project?  And
> then we can point to a specific tag/hash of that repo.  If we don't hear
> anything from them after a few weeks then we could consider your fork,
> but if we can have the hooks upstream from the start it's less
> maintenance.
> 

I've done that, and submitted fixes for two problems.

If that doesn't work out, perhaps some setup on our local sourceware 
forge owned by the gdb project rather than my personal github account 
would be better.  Not sure how to get that setup though, I haven't used 
it yet.

Thanks,
- Tom

> Simon