newsletter #10

newsletter #10

[Mathieu Lacage]

hi all,

http://gccnews.chatta.us/

or, for your conveniance:

gcc release

Mark Mitchell sent in a gcc status summary[1] for gcc 3.4.2. He then
declared the 3.4.2 branch frozen[2], released[3] 3.4.2-RC1, then
3.4.2-RC2 and then 3.4.2[4].
Mark also sent in a gcc 3.5 status summary[5] which, not surprisingly,
triggered a few more or less angry mails about projects dropped in or
out of the 3.5 release. It is clear to everyone that if the 3.5 release
follows the path set forward by Mark, it will suffer from the
dot-zero[6] syndrome. Although some people are very unhappy by this and
would prefer to delay the release for a few months to try to make gcc
3.5 really better all the time than previous releases, this is unlikely
to happen (There is a strong argument against such a move since it is
well known that delaying a release to work more on it often,
strangely-enough, entices engineers to write more code and thus more
bugs. hrm. I certainly never do that, Do I ?).

gcc development

Daniel Berlin spent a bit of time gathering information on the current
register allocator work and on future directions. He summarized the
status of the current architecture in this mail[7]. Another thread was
started by Caroline Tice to provide a bit more information[8] on the
topic.
More compiler warnings are always good. It looks like the OpenBSD folks
added the sentinel (a [.] checker that can verify that varargs functions
such as execl(cmd, arg1, arg2, NULL) are indeed null-terminated (when
the arguments are constant)) attribute to their 2.95 gcc. Marc Espie
posted the patch[9] implementing this for 2.95 gccs. Hopefully, this
will get in someday.
Work on new optimization passes for gcc has not stalled despite the
tentative release schedule. Revital Eres sent a new version of his
Variable Exansion Optiomization[10] pass.

documentation

Anthony Green posted a link to the original design document[11] for gcj
by Cygnus back in the old days.
Thanks to Joe Buck and Daniel Berlin, the gcc summit proceedings have
now been split into separate papers and are hosted on gcc.gnu.org[12].

libstdc++

Contributed by jcopenha (jcopenha at typedef dot org)
 The libstdc++ mailing list is a rather low volume list. There were some
patches submitted this week. The first[13] being a fix for do_get_time
in time_get::get_time to handle leap seconds correctly. To stay in
alignment with C99 if _GLIBCXX_USE_C99 is defined the valid values for
seconds are 0 - 60, if using C89 0 - 61. An interesting paper on the
history of leap seconds[14] was also posted.
 The second was a fix[15] for PR libstdc++/17215, __basic_fileunderlying
fclose call. This meant the user never new if __basic_fileto soon, if
for instance fclose failed with EINTR. It was decided to mimic the read
and write behavior and loop on EINTR.

[0] http://gcc.gnu.org/ml/gcc/2004-08/msg01117.html
[1] http://gcc.gnu.org/ml/gcc/2004-08/msg01412.html
[2] http://gcc.gnu.org/ml/gcc/2004-08/msg01415.html
[3] http://gcc.gnu.org/ml/gcc/2004-09/msg00348.html
[4] http://gcc.gnu.org/ml/gcc/2004-08/msg01418.html
[5] http://gcc.gnu.org/ml/gcc/2004-08/msg01446.html
[6] http://gcc.gnu.org/ml/gcc/2004-08/msg00900.html
[7] http://gcc.gnu.org/ml/gcc/2004-08/msg00904.html
[8] http://gcc.gnu.org/ml/gcc/2004-08/msg01105.html
[9] http://gcc.gnu.org/ml/gcc/2004-08/msg01050.html
[10] http://www.spindazzle.org/cygnus/design.html
[11] ftp://gcc.gnu.org/pub/gcc/summit/
[12] http://gcc.gnu.org/ml/libstdc++/2004-08/msg00215.html
[13] http://www.cl.cam.ac.uk/~mgk25/time/metrologia-leapsecond.pdf
[14] http://gcc.gnu.org/ml/libstdc++/2004-08/msg00220.html



Mathieu
-- 
Mathieu Lacage <mathieu.lacage@sophia.inria.fr>

Re: newsletter #10

[Kaveh R. Ghazi]

 > More compiler warnings are always good. It looks like the OpenBSD
 > folks added the sentinel (a [.] checker that can verify that
 > varargs functions such as execl(cmd, arg1, arg2, NULL) are indeed
 > null-terminated (when the arguments are constant)) attribute to
 > their 2.95 gcc.  Marc Espie posted the patch[9] implementing this
 > for 2.95 gccs. Hopefully, this will get in someday.

This is done.

I implemented something slightly more encompassing for mainline based
on feedback from a long discussion here:
http://gcc.gnu.org/ml/gcc/2004-08/msg01068.html
(The sentinel-value argument from my proposal was not included.)

My patch went in just before the stage3 deadline so the next release
of GCC will have it:
http://gcc.gnu.org/ml/gcc-patches/2004-09/msg00505.html

The difference from Marc Espie's original implementation for 2.95 is
that mine also can check functions like `execle' where the sentinel is
not the last variadic argument.

		--Kaveh
--
Kaveh R. Ghazi			ghazi@caip.rutgers.edu

Re: newsletter #10

[Ranjit Mathew]

Mathieu Lacage wrote:
> Work on new optimization passes for gcc has not stalled despite the
> tentative release schedule. Revital Eres sent a new version of his
> Variable Exansion Optiomization[10] pass.
           ^^^----------^--- Typos.

IMHO, also important additions were Stack Slot Coalescing:

  http://gcc.gnu.org/ml/gcc-patches/2004-09/msg00333.html

and Loop Interchanging (on the LNO branch only though):

  http://gcc.gnu.org/ml/gcc-patches/2004-09/msg00162.html

And an inordinate amount of time was also spent on discussing
the need to have Ada tested by default due to GNAT version
incompatibilities.

Jan Hubicka has been hunting memory leaks and trying to
rein in memory usage.

Nathan Sidwell has been gcc_assert()-ifying the code base.

An unusual amount of bootstrap breakage on various
platforms...

Daniel Berlin has set up a GCC Wiki at
http://dberlin.org/gccwiki/ ...

Et cetera.

Ranjit.

-- 
Ranjit Mathew          Email: rmathew AT gmail DOT com

Bangalore, INDIA.      Web: http://ranjitmathew.tripod.com/

Re: newsletter #10

[Gabriel Paubert]

On Wed, Sep 08, 2004 at 02:20:58PM +0200, Mathieu Lacage wrote:

[snipped]
> Mark Mitchell sent in a gcc status summary[1] for gcc 3.4.2. He then
> declared the 3.4.2 branch frozen[2], released[3] 3.4.2-RC1, then
> 3.4.2-RC2 and then 3.4.2[4].
> Mark also sent in a gcc 3.5 status summary[5] which, not surprisingly,
[snipped]
> patches submitted this week. The first[13] being a fix for do_get_time
> in time_get::get_time to handle leap seconds correctly. To stay in
> alignment with C99 if _GLIBCXX_USE_C99 is defined the valid values for
> seconds are 0 - 60, if using C89 0 - 61. An interesting paper on the
> history of leap seconds[14] was also posted.
>  The second was a fix[15] for PR libstdc++/17215, __basic_fileunderlying
> fclose call. This meant the user never new if __basic_fileto soon, if
> for instance fclose failed with EINTR. It was decided to mimic the read
> and write behavior and loop on EINTR.
> 
> [0] http://gcc.gnu.org/ml/gcc/2004-08/msg01117.html
> [1] http://gcc.gnu.org/ml/gcc/2004-08/msg01412.html
> [2] http://gcc.gnu.org/ml/gcc/2004-08/msg01415.html
> [3] http://gcc.gnu.org/ml/gcc/2004-09/msg00348.html
> [4] http://gcc.gnu.org/ml/gcc/2004-08/msg01418.html
> [5] http://gcc.gnu.org/ml/gcc/2004-08/msg01446.html
> [6] http://gcc.gnu.org/ml/gcc/2004-08/msg00900.html
> [7] http://gcc.gnu.org/ml/gcc/2004-08/msg00904.html
> [8] http://gcc.gnu.org/ml/gcc/2004-08/msg01105.html
> [9] http://gcc.gnu.org/ml/gcc/2004-08/msg01050.html
> [10] http://www.spindazzle.org/cygnus/design.html
> [11] ftp://gcc.gnu.org/pub/gcc/summit/
> [12] http://gcc.gnu.org/ml/libstdc++/2004-08/msg00215.html
> [13] http://www.cl.cam.ac.uk/~mgk25/time/metrologia-leapsecond.pdf
> [14] http://gcc.gnu.org/ml/libstdc++/2004-08/msg00220.html

Very good job, but there is an off by one error: you count the 
references starting with [1] in the text but with [0] in the
list of links. Not very serious but probably easy to correct
for the next release.

	Regards,
	Gabriel

RE: newsletter #10

[Dave Korn]

> -----Original Message-----
> From: gcc-owner On Behalf Of Gabriel Paubert
> Sent: 08 September 2004 14:59

> On Wed, Sep 08, 2004 at 02:20:58PM +0200, Mathieu Lacage wrote:
> 
> [snipped]
> > Mark Mitchell sent in a gcc status summary[1] for gcc 3.4.2. He then
> > declared the 3.4.2 branch frozen[2], released[3] 3.4.2-RC1, then
> > 3.4.2-RC2 and then 3.4.2[4].

> > [0] http://gcc.gnu.org/ml/gcc/2004-08/msg01117.html
> > [1] http://gcc.gnu.org/ml/gcc/2004-08/msg01412.html

> Very good job, but there is an off by one error: you count the 
> references starting with [1] in the text but with [0] in the
> list of links. Not very serious but probably easy to correct
> for the next release.

  Only if it counts as a regression...... otherwise it has to wait for the
next version!

    cheers, 
      DaveK
-- 
Can't think of a witty .sigline today....