Re: fix audit testcase with IE TLS disabled

[Alexandre Oliva <aoliva at redhat dot com>]

On Mar 19, 2006, Roland McGrath <roland@redhat.com> wrote:

> It only took me a few minutes to write a test case that
> readily demonstrates the bug, based directly on your description of the
> scenario.

Clever!  I spent a lot of time trying to do that with your malloc hook
suggestion, but I failed miserably, and I still don't quite understand
why nothing that I tried would work :-(

> I think this would have had a greatly better chance of getting dealt
> with sooner if you had included a patch to add a test case like the
> one I've just put in.

Believe me, I tried.  Several times.  I just failed to think of trying
the simple approach that you did, because I didn't realize a calloc
definition in the main executable would have been used by the dynamic
loader at that point.

> Finally, the best way we have for making sure things don't fall on
> the floor is bugzilla; I had your message sitting for months in a
> mail folder knowing I had to look at it one day, but that's not
> reliable.

Understood.  I suppose I should file a bugzilla about the two other
pending patches related with TLS I have, or do you think it at this
point you're close enough to getting to them that the bug reports
would just add overhead instead of helping?

Thanks again.  I'll keep trying to remember to use standard ChangeLog
headers, if what you mean is my use of `from' instead of the date.
Odds that I succeed are higher nowadays.  If anyone is interested, let
me explain why I used to consistently fail at this.  For whoever
isn't, it's time to skip to the next message :-)


Knowing that ChangeLog entries are not supposed to be patches, but
needing a convenient way to deal with them, I came up with two
scripts, clcleanup, that transforms a patch that includes changes to
ChangeLog files into a patch like the one I post, and cl2patch, that
turns ChangeLog entries formatted like this back to patches that apply
to the top of current ChangeLog files.

These scripts are part of GNU CVS Utilities, so far unreleased but
with a CVS repository at gnu.org (details at my home page, URL
below).

More recently, I started using quilt for patch management, which
actually required me to change my practices as to patch formatting,
since the use of Index: for ChangeLog text entries gets its latest
releases very confused.  Since I adopted quilt, I stopped editing
ChangeLog files directly, using the quilt headers for this purpose, so
now I will hopefully remember to use a date instead of `from´.

That still feels wrong to me, since I'd post a patch with today's date
to have it applied most often at a different date, at which point I
understand the ChangeLog date should be changed to the patch
application date (mind you, something that cl2patch can actually fix
for you :-)

clcommit is another useful script in this collection, in that it can
extract the changes from the ChangeLog files you intend to check in
and use those as the check in message.

-- 
Alexandre Oliva         http://www.lsd.ic.unicamp.br/~oliva/
Secretary for FSF Latin America        http://www.fsfla.org/
Red Hat Compiler Engineer   aoliva@{redhat.com, gcc.gnu.org}
Free Software Evangelist  oliva@{lsd.ic.unicamp.br, gnu.org}