This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: stackable abilist files


The main point of the abilist stuff is that it is 100% mechanical.
There's no possibility for human error in editing the data, because
no human ever edits the data.  A human only looks at some changes
and says "yes, we meant that" before committing.  We should not do
anything that introduces new possibilities for error.

One reason the existing scheme never used sysdeps files is that this
would introduce a new potential kind of error.  That is, a change to
the set of sysdeps directories used would mean suddenly checking
against different data for the same configuration tuple.  But if the
check is mandatory and every configuration must have complete and
correct data to pass 'make check', then it seems implausible that
such an error would slip through.

Much as I hate to let some good awk scripts go to waste, it no
longer seems worthwhile to do the merged files.  All the abilist
files together are about 80k.  The files for an individual
configuration would be smaller, so take that as an upper bound.
Assuming we have 10 working configurations (random plausible
number), that's 800k of text that compresses well.  For comparison,
we have over 6mb of ChangeLog files.  It just doesn't seem like a
big deal any more.  (I'm not even really sure it was a big deal
ten years ago, but it seemed so at the time.)

Another feature of the merged files is that gratuitous differences
between configurations pop out quickly to the eye.  It's certainly
good to avoid such differences creeping in unnoticed.  But we could
easily enough write some scripts to display the information in all
the files in a way that makes the differences easy to notice.

I don't think we really need the configure option any more.  It's
easy enough to provide an ad hoc makefile variable that people can
set on the make command line to disable the check if anyone cares
to.


Thanks,
Roland


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]