egcs snapshot failures building on cygwin-1.1.0

[Tim Prince]

Somehow, cgf's response to my posting had removed permission to
respond, so I am writing a new message as a response.

The problems I mention are not cygwin problems, although I am
concerned when egcs-testsuite results produced by dejagnu
changed with the new cygwin version.  I suspect that the current
ones are more accurate (gcc reported too few failures before),
but expect crashes when trying to test g++, probably as a result
of one of the many snapshot g++ crashes.  gcc-2.95.2 (and g77)
are working like a charm.

I do have some concern, considering that they are talking about
a gcc-3.0 release, while the state of the current snapshots is
so bad, particularly on cygwin.  The last snapshots which built
at all on cygwin did so only without optimization.  Mumit Khan
submitted a patch last week to fix a place where they broke the
cygwin configuration, but they haven't incorporated it yet.  I
had been using my own version of that one for weeks.  Earlier
snapshots were working fairly well on the old cygwin, except
that several of the optimizations I want are in the -ffast-math
collection now, and that has been buggy for months.  Even if
they focus on linux and don't care about cygwin, they will be
needing to fix some of these problems.  I'd go find out which
ones I could show occur both on cygwin and linux if it would do
any good.  Sometimes, we might be able to take the cygwin
version of the preprocessed file and produce the same ICE under
linux, if that means anything to anyone.

I'm neutral on the question of making special cygwin features
work in gcc, but adamant about wanting the standard linux ones
to work as far as possible.

One place where I don't know whether to blame myself is that
makeinfo has been broken because there are some \r characters in
the original snapshot .texi's, and cygwin patch has been adding
more.  I haven't checked this fully on 1.1.0.  All my mounts are
binary.  I am using the cwilson makeinfo, which I think is
working now, unless it ought to accept \r's.

I also must re-check the question of whether sharutils can be
made to pass its test suite without playing tricks with mount
options.  Shouldn't patch and sharutils work with binary mounts?
If they don't, is it due to a mis-configuration in their build?


--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com