Re: canonicalize-lgpl bug

[Eric Blake]

Jim Meyering <jim <at> meyering.net> writes:

> >       [3/11] canonicalize: don't lose errno
> > glibc still has a bug in realpath/c_f_n where errno could be inadvertently
> > changed by a call to free() during an error return, but canonicalize-lgpl 
was
> > immune, and now canonicalize is fixed.  I guess I'll have to file a glibc 
bug
> > report for the gnulib->glibc syncing (patch 9 gets the glibc->gnulib 
syncing)
> 
> To ease future glibc/gnulib syncing, it'd be better not to change
> __set_errno (...) to errno = ...  No?

I guess I was a bit confusing in my description.  canonicalize-lgpl and glibc 
are in sync (well, patched to be in sync), and in that file, I continued to 
keep __set_errno.  canonicalize (GPL) adds a new interface, c_f_mode, and given 
our track record of convincing Ulrich to add new interfaces to glibc, it's 
doubtful that we'll be able to push any of canonicalize.c back to glibc.  
Besides, canonicalize.c has already made many other breaks from when it 
originally forked from glibc, such as renaming 'rpath' to 'rname'.  So this 
particular __set_errno=>errno cleanup is in line with gnulib owning the file.

> 
> Also, if you do make a mechanical change like that, it's easier
> on reviewers and general maintenance to keep that in a change-set
> separate from any delta that makes a semantic change, like your
> "don't lose errno" fix does.

OK, I'll split that patch into two.

> 
> I've reached this point in reading the patches.
> So far they look fine.
> I will read the remainder, and test tomorrow.

I'll try and rebase my series before then.

> BTW, I appreciate these commentaries.
> Have you considered adding something like that to each commit log?

Will do as part of my rebase.

-- 
Eric Blake