Re: [IPCop-devel] Testing gcc-4.4.6 upgrade

[Gilles E. <g....@fr...>]

----- Original Message ----- 
From: "Olaf Westrik" <wei...@ip...>
To: "Gilles Espinasse" <g....@fr...>
Cc: "IPCOP devel" <ipc...@li...>
Sent: Saturday, May 14, 2011 6:12 PM
Subject: Re: [IPCop-devel] Testing gcc-4.4.6 upgrade


> On 2011-05-11 19:55, Gilles Espinasse wrote:
>
> > Is it the time to update gcc?
>
> Not sure, other than less errors when testing, what would be the benefit
> for us?
>
>
> Olaf

We could prefer gcc-4.4.6 than gcc-4.4.5 as more bugs should have been fixed
than include in the last release.
In true life, when you are hit by a bug, it may require some hard time
before to know that the reason is in the compiler.

I agree there is no urgency to upgrade. I will keep that change on one
machine and see if the other machines behave differently.
I had other changes coupled with 4.4.6 in mind. If you disable static lib on
gcc, our installer and gdb build will fail because of TEXTREL issue.
Adding -fPIC on both solve those issues (the produced code is a bit bigger).
I haven't yet checked if every executable is PIE compiled and should do
that.

Actually, we know of our build bug and we have absolutly no idea where it
does come from.
That's the know issue where our build sometime and very unfrequently fail
for unidentified reason.
Once during ./make.sh clean && ./make.sh build, e2fsprogs fail to build
because pkg-config was not found. And that was true, pkg-config have not be
build _that time_ for the reason know from our bug only. That's a bit scary
but as we can't reliabily reproduce, we can't debug.

For that reason, I have tried to upgrade make to 3.82. That compile and is
unlikely to produce different binaries than 3.81 (I haven't yet checked
that). I will see with usage if our build bug is still there. For 3.82,  I
have one patch borrowed from fedora that fix 3 bugs, fedora has more
patches, maybe 2 or 3 other may fix real issues for us. Gentoo has 2 patches
that do mostly the same as the one from fedora I have compiled, as all of
that is fixes from upstream cvs.

There is one oddity in 3.82 vs 3.81 in the log produced
diff -ur log_i486-3.81/02_base/rsyslog-5.6.5 log_i486/02_base/rsyslog-5.6.5
--- log_i486-3.81/02_base/rsyslog-5.6.5 2011-05-13 17:51:22.000000000 +0200
+++ log_i486-3.82/02_base/rsyslog-5.6.5 2011-05-13 23:40:50.000000000 +0200
@@ -7,9 +7,9 @@
 Hunk #2 succeeded at 221 (offset -2 lines).
 patching file runtime/wtp.c
 cd /usr/src/rsyslog-5.6.5 && ./configure --prefix=/usr \
-     --exec-prefix=/usr \
-     --sysconfdir=/etc \
-     --localstatedir=/var
+    --exec-prefix=/usr \
+    --sysconfdir=/etc \
+    --localstatedir=/var
You may see nothing different here if your mailer does not render tab as 8
char (I suspect mine will replace tab with space).
With 3.82, there is one tab less for the characters after the first
continuation character : the \
After first configure line, for the next options, there is 5 tab with 3.81
and 4 with 3.82
So the log diff between 3.81 and 3.82 is big (1.3 MB) and hard to read.
I am unsure this is a bug. In makefile, the tab mean this is a command. On
the first line, that tab did not appear on log, so it may be normal make
suppress one tab on the lines that follow the first.

I now know too why the cairo test failure count is so big, but I don't know
yet the fix : the fonts are not found by cairo during the tests.
That look like https://bugs.freedesktop.org/show_bug.cgi?id=18077
Despite reading cairo test/README, I have no idea where to change something.
fc-list know the installed fonts.

Gilles