Hi,

Sorry for the delay!

rixed@happyleptic.org skribis:

> I'm refering to the ensure-writable-dir function.
> I've stumbled upon this problem recently : I have a program that's suid root
> but calls guile. The guile compiler then creates some directories in
> .cache/guile but then check (with access()) that he can write in there, which
> he can't since access revoke the effective uid for the caller uid. We thus have
> this situation: the compiler creates a bunch of directories then complains he
> cannot write in them.
>
> It's not obvious to me why the compiler should ensure a directory is writable
> just to throw an error. Wouldn't it be better to just call opens and writes
> and let these fails and report these more acurate errors instead (or, in my
> case, let them succeed) ?

Agreed, not to mention time-of-check-to-time-of-use-errors.

We basically want ‘mkdir -p’, but we can even omit the stat(2) call upon
EEXIST, because if DIR doesn’t point to a directory, the error will be
caught soon after anyway.

I’ll install the following patch if there’s no objection: