bug#33567: Syntactic fontification of diff hunks

[Juri Linkov]

>> For more safety, I propose to set a new buffer-local variable
>> `diff-default-directory' by such commands as diff, diff-backup,
>> dired-diff, dired-backup-diff.  The existence of such variable
>> should guarantee that the referenced files really exist.
>> This variable will be like `diff-vc-backend' that says that
>> the diff-mode buffer is created by the VCS command.
>> Then anyone who want to visit a diff file in another directory,
>> could add it to the first line:
>>
>> -*- mode: diff-mode; diff-default-directory: "..." -*-
>
> I'm not sure this is a step in the right direction.  What is the
> advantage of having a separate variable?  How is it "safer"?

When this special variable is set by a diff command, it's safe to assume
that the files referenced from the diff buffer really exist, so it's
safe to try reading them.  I don't want a patch in a mail attachment
to try reading files mentioned in the patch attachment.

>> >> > Also, if the diffs are from Git, they begin with a/, b/, etc. dummy
>> >> > directories, which usually don't exist in the file system.
>> >>
>> >> This is not a problem because diff-find-file-name used in the patch
>> >> strips such a/, b/ prefixes to get the existing file name.
>> >
>> > Not in my testing, but maybe I tried in the wrong Emacs version.  Is
>> > this feature new with Emacs 27?
>>
>> For testing better try to eval e.g. `(diff-find-file-name nil t)'
>> on a hunk in a diff-mode buffer created by git.
>
> I did, but I guess this must be done inside the repository to work,
> does it?  If I put the output of "git diff" on a file in some
> arbitrary directory, then visit that file and evaluate
> (diff-find-file-name nil t), I get nil.

Yes, it finds only the existing files inside the repository.

>> >> +For working revisions get highlighting according to the working
>> >> +copy of the file.
>> >
>> > I don't understand the significance of this comment.  If you want to
>> > say that the produced highlighting might be wrong if the working
>> > version has changed since it was compared, then let's say that
>> > explicitly.
>>
>> This means that working revisions can't be extracted from the repository.
>> Until committed, they reside in files that are visited with find-file.
>
> We need to describe the implications of that to the users.  Does the
> following text capture the issue?
>
>   For diffs against the working-tree version of a file, the
>   highlighting is based on the current file contents, which could be
>   different from the contents when the diffs were taken.  In such
>   cases, the produced highlighting might be wrong.

Such problem is very rare because highlighting is added usually
immediately after creating a diff.  When the file contents changes,
there is no highlighting at all - it verifies if text of the hunk
exist in the file, so highlighting never is wrong.

>> >> +If t, additionally to trying to use a version control system to get
>> >> +old revisions for fontification, also try to get fontification based
>> >> +on existing files, and on failure get fontification from hunk alone."
>> >
>> > What is the difference between using a VCS to get old revisions, and
>> > using existing files?
>>
>> This means that when a diff-mode buffer is not created by a VCS,
>> then it tries to read files with find-file.
>
> If so, I suggest the following wording:
>
>   If t, try to infer fontification from the compared files, if they
>   exist in the filesystem, when accessing their contents via VC
>   fails.

Will add in the final patch.

>> > Also, does it mean `vc' will not fall back to `hunk-only'?  Why not?
>>
>> Actually, it already falls back to `hunk-only', this is what
>> "on failure get fontification from hunk alone." tries to say.
>
> There's no such text in the description of 'vc', only in the
> description of t, which is why I asked.

Maybe then better to add text common for all cases, e.g.

"If some method fails, get fontification from hunk alone."