This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: conformtest: Support xfail markers on individual assertions [committed]
- From: Joseph Myers <joseph at codesourcery dot com>
- To: Roland McGrath <roland at hack dot frob dot com>
- Cc: <libc-alpha at sourceware dot org>
- Date: Fri, 19 Jun 2015 20:31:07 +0000
- Subject: Re: conformtest: Support xfail markers on individual assertions [committed]
- Authentication-results: sourceware.org; auth=none
- References: <alpine dot DEB dot 2 dot 10 dot 1506192006270 dot 1466 at digraph dot polyomino dot org dot uk> <20150619201745 dot EA76D2C3B31 at topped-with-meat dot com>
On Fri, 19 Jun 2015, Roland McGrath wrote:
> I like the intent. I noticed that almost all of the instances in the -data
> files use a consistent convention for the comment:
> // Bug NNNNN: ...
>
> My first impulse was to ask that the convention be stated somewhere and
> followed uniformly. But it occurs to me it would be nice to have this be
> more formal than a comment convention. How about instead of:
> // Bug NNNNN: explanatory comment
> xfail-foo ...
> we make it:
> // explanatory comment
> xfail {NNNNN} foo ...
>
> This makes it completely explicit that you can't have an XFAIL without
> having a bug filed for it. It also makes it marginally easier than a
> comment convention to have scripts that do something or other with the bug
> number.
There are cases where it's not clear having bugs filed is useful. (For
example: XPG4 specifies st_blksize and st_blocks as of type long, where
subsequent standards change them to blksize_t and blkcnt_t. So when
properly reviewed for XPG4, the data should reflect this - but should also
have xfail markers, given that those types aren't always long. And as
that's simply a limitation in the level of support for an old standard -
much like we have the varargs.h tests XFAILed at makefile level - I don't
see a use in a bug for it.)
I did wonder about making the syntax include somewhere for a condition
under which the xfail applies - to allow for architecture-specific xfails
(hard-to-fix issues with types can often be architecture-specific) - but
that didn't seem necessary at this point.
--
Joseph S. Myers
joseph@codesourcery.com