bug#23723: patch-shebang phase breaks symlinks

[Jelle Licht]

Ludovic Courtès <address@hidden> writes:

> Jelle Licht <address@hidden> skribis:
>
>> Ludovic Courtès <address@hidden> writes:
>>> Hello!
>>>
>>> Jelle Licht <address@hidden> skribis:
>>>
>>>> It seems that the patch-shebang functionality does not deal gracefully
>>>> with symlinks: it just overwrites them!
>>>>
>>>> After struggling somewhat with getting the recently packaged node 6.0.0
>>>> to behave, I found out that `patch-shebang' in (guix build
>>>> gnu-build-system) does not work properly on symlinks.
>>>
>>> There’s ‘patch-shebangs’ (plural) in this file, but it explicitly
>>> touches only regular files (see ‘list-of-files’).
>>>
>>
>> It seems I made a mistake when writing the bug report; I am talking
>> about the `patch-shebang' defined in (guix build utils). My apologies.
>>
>> Also, seeing as my experience with the stat utility and similarly styled
>> programming libraries was lacking, I decided to play around with the
>> definition of `list-of-files': It actually does include symlinks, as
>> (stat:type (stat "some-symlinked-file")) gives us a plain old 'regular.
>> Looking into this a bit more, it seems that calling `stat' gives the
>> exact same results on both the linked-to-file and the symlink to that
>> file.
>>
>> For the particular problem I ran into to be fixed, it is imperative that
>> `list-of-files' of `patch-shebangs' includes the symlink; it does after
>> all need to be patched. The way this patching currently happens just
>> clobbers symlinks.
>
> My bad, indeed, ‘list-of-files’ should use ‘lstat’ instead of ‘stat’.

This would be one way of fixing this bug. I'd rather see that
`patch-shebang' in (guix build utils) checks for symlinks, and if so,
patches the actual file instead of the symlink. This is the approach I
currently use in my tree to use node 6.0. Would there be any downside to
this approach?

>
> I think a patch like attached should solve the problem.  WDYT?
>
> We can apply it to core-updates-next if that’s fine with you.
>
> Thanks,
> Ludo’.

If this is the approach you'd rather follow, that is okay with me as
well. I just think that a phase that transparently patches all files in
bin/sbin, whether they are actual files or symlinks, would be more
useful.

Greetings,
- Jelle