# 04:12:26 |
tzarc |
joins #crosstool-ng |
# 04:13:35 |
tzarc |
evening folks |
# 04:13:56 |
tzarc |
don't suppose anyone knows of a solution for: configure: error: Link tests are not allowed after GCC_NO_EXECUTABLES. ? |
# 04:14:53 |
tzarc |
trying x86_64 -> arm, but it fails during the cc_for_host stage |
# 08:07:37 |
arekinath |
quits : Read error: Operation timed out |
# 08:11:17 |
arekinath |
joins #crosstool-ng |
# 08:28:53 |
kos_tom |
tzarc: Yann was having the same problem yesterday. I'd guess that if he has time today, he'll look into this. |
# 08:30:03 |
tzarc |
cool. had a look at the logs and saw it, attempting to work my way around it with script modification |
# 08:30:17 |
tzarc |
will feed it back if I can get it sorted I guess |
# 08:59:21 |
tzarc |
mmm, well it seems to be working now |
# 08:59:28 |
tzarc |
kinda heavy-handed approach though |
# 09:49:43 |
arekinath |
quits : Ping timeout: 250 seconds |
# 09:50:56 |
arekinath |
joins #crosstool-ng |
# 09:50:56 |
arekinath |
quits : Changing host |
# 09:50:56 |
arekinath |
joins #crosstool-ng |
# 12:40:03 |
y_morin |
joins #crosstool-ng |
# 13:10:27 |
y_morin |
Hello! Any idea how the new armhf tuples should be interpreted? |
# 13:10:45 |
y_morin |
Eg. arm-unknown-linux-gnueabihf is valid |
# 13:10:59 |
y_morin |
but what about arm-unknown-linux-uclibcgnueabihf ? |
# 13:15:08 |
tzarc |
that doesn't really make sense, does it? |
# 13:15:16 |
tzarc |
it'd be uclibchf, no? |
# 13:15:27 |
tzarc |
gnu matches glibc? |
# 13:16:21 |
y_morin |
tzarc: not really, in fact. The proper EABI tuple for ARM and uclibc is arm-unknown-linux-uclibcgnueabi |
# 13:16:54 |
y_morin |
tzarc: GNU is everywhere... ;-] |
# 13:16:57 |
kos_tom |
just curious, how do you know which tuple is correct for a particular configuration? is there a specification of existing tuples somewhere? |
# 13:17:08 |
tzarc |
hmmmm |
# 13:17:22 |
y_morin |
kos_tom: the config.guess and config.sub scripts are the soecs, I gues... :-/ |
# 13:17:28 |
y_morin |
*specs... |
# 13:17:56 |
y_morin |
http://git.savannah.gnu.org/gitweb/?p=config.git |
# 13:18:36 |
kos_tom |
urgh, what a spec :-( |
# 13:18:43 |
y_morin |
Yep... |
# 13:35:26 |
tzarc |
I was looking at sorting out the "Link tests are not allowed after GCC_NO_EXECUTABLES." error earlier today |
# 13:36:26 |
tzarc |
for LE arm, just need to suppress "-EL" in do_cc_backend() |
# 13:36:33 |
ccole |
joins #crosstool-ng |
# 13:49:26 |
y_morin |
tzarc: -EL is set in scripts/functions on line 980. |
# 13:49:47 |
tzarc |
yep, but stage 2 fails if you remove it there |
# 13:49:56 |
y_morin |
-EL is a valid ld flag. Alas, the fscking autostuff inisist on callin 'gcc' as the linker. Gah... |
# 13:50:09 |
tzarc |
mmmm |
# 13:50:14 |
y_morin |
tzarc: that's even weirder. stage-2 builds with -EL, final does not. WTF? |
# 13:50:27 |
tzarc |
I know, eh |
# 13:50:29 |
y_morin |
:-) |
# 13:50:30 |
tzarc |
makes no sense |
# 16:24:36 |
y_morin |
tzarc: thank you for the -EL hint. I have a patch here that I am testing, and preliminary tests seem to be conclusive! Thanks! |
# 16:25:03 |
y_morin |
kos_tom: seems we are on the verge of fixing the gcc-4.7 build. |
# 16:25:20 |
kos_tom |
y_morin: ah ah? |
# 16:25:43 |
kos_tom |
I don't *need* it myself, but I guess being able to build 4.7 in ct-ng will be useful at some point in the future :) |
# 16:25:58 |
y_morin |
kos_tom: yep, tzarc said that removing -EL in LDFLAGS made final gcc-4.7 build OK, but that core pass-2 needed it. |
# 16:26:12 |
y_morin |
kos_tom: So I replace -EL witj -WL,-EL, and it builds! :-) |
# 16:27:18 |
y_morin |
Tss... keyboard-dyslexia... |
# 16:27:33 |
kos_tom |
doesn't this sounds a bit hackish? is the fundamental reason understood? |
# 16:28:17 |
y_morin |
kos_tom: not yet fully understood, indeed, but at least we at least have a direction to follow, now! :-) |
# 16:30:14 |
kos_tom |
just curious, do you know why the Linaro guys have moved libstdc++, libgcc_so and other gcc libraries outside of the sysroot since 2012.03 ? |
# 16:30:50 |
y_morin |
In there pre-built toolchain? |
# 16:30:54 |
y_morin |
*their |
# 16:31:35 |
kos_tom |
yes |
# 16:32:10 |
y_morin |
I did not look at their toolchain, but I supposed they moved it to the libexec sub-dir, right? |
# 16:33:57 |
kos_tom |
I guess it's related to the "Subject: GCC support libraries and sysroot |
# 16:33:58 |
kos_tom |
" |
# 16:33:59 |
kos_tom |
thread |
# 16:34:01 |
kos_tom |
on the crossgcc list |
# 16:37:51 |
y_morin |
kos_tom: OK, I remember that thread. The fact that those libs are in the sysroot is not the "standard" layout. |
# 16:38:34 |
y_morin |
kos_tom: the canonical layout is to have those "support" libs in their own dir, and the sysroot would only contain the C library libs |
# 16:38:53 |
y_morin |
Which would make it easy to replace the sysroot with an alternate one. |
# 16:39:40 |
y_morin |
In the beginings of ct-ng, that "canonical" layout was not working (probably because of a bug in ct-ng), but I noticed that: |
# 16:40:03 |
y_morin |
- having all the taget libs in the sysroot made the build work (that not a reason for doing so, though) |
# 16:40:39 |
y_morin |
- I found it stupid to have target libs outside the sysroot |
# 16:41:09 |
kos_tom |
ok, so it doesn't sound crazy for BR to support the case where the support libs are outside the sysroot, right? |
# 16:42:12 |
y_morin |
kos_tom: I'd say ct-ng is not doing things properly from an "academic" point-of-view. So yes, BR (like other build systems) should be able to cope with the fact that those libs could be {any,some}where else. |
# 16:42:53 |
kos_tom |
basically, I could do a gcc -print-file-name=libstdc++.a and assume that this is the "support libraries" directory, correct ? |
# 16:42:57 |
y_morin |
kos_tom: the best bet would be to ask gcc with: gcc -print-file-name=... |
# 16:43:20 |
kos_tom |
so I do one -print-file-name=libc.a to get the sysroot, and another -print-file-name=libstdc++.a to get the support lib directory |
# 16:43:26 |
y_morin |
kos_tom: I'd do it for every needed lib, and copy the pointed-to lib |
# 16:43:28 |
kos_tom |
and then I can merge both into the BR sysroot |
# 16:43:41 |
kos_tom |
well, for the sysroot, we don't copy lib-per-lib, we copy the whole thing |
# 16:43:51 |
kos_tom |
we do lib-per-lib copy for the target |
# 16:44:51 |
y_morin |
kos_tom: remind me why you need to copy the sysroot in the first place? Is that because you override its location with --sysroot=... ? |
# 16:45:04 |
y_morin |
At the sysroot you point to is the staging area? |
# 16:45:11 |
y_morin |
*And... |
# 16:45:38 |
kos_tom |
right |
# 16:48:51 |
y_morin |
kos_tom: they've gone one step further in 2012-04: the libc is not even installed in the sysroot... :-/ |
# 16:48:59 |
y_morin |
bin/arm-linux-gnueabi-gcc -print-file-name=libc.so |
# 16:49:07 |
y_morin |
gcc-linaro-arm-linux-gnueabi-2012.04-20120426_linux/bin/../arm-linux-gnueabi/libc/usr/lib/arm-linux-gnueabi/libc.so |
# 16:49:15 |
y_morin |
(stripped leading path) |
# 16:49:31 |
y_morin |
kos_tom: notice where it finds the libc.so? |
# 16:49:40 |
y_morin |
yep, it's a linker script, but anyway... |
# 16:50:29 |
y_morin |
The real libc.so is just side-by-side with this linker script... |
# 16:50:50 |
y_morin |
And there is no sysroot anymore... :-( |
# 19:28:02 |
ben1066 |
quits : Ping timeout: 246 seconds |
# 19:29:15 |
ben1066 |
joins #crosstool-ng |
# 20:50:55 |
ben1066_ |
joins #crosstool-ng |
# 20:53:52 |
ben1066 |
quits : Ping timeout: 276 seconds |
# 22:44:14 |
y_morin |
kos_tom: OK, I have a patch locally that uses savedefconfig, now! |
# 22:44:46 |
y_morin |
kos_tom: I have other pending changes to test, and I'll push all that tomorrow, after some testing. |
# 22:45:04 |
y_morin |
kos_tom: and indeed, savedefconfig makes for much smaller .config files! Yeah! \o/ |
# 23:06:47 |
y_morin |
quits : Quit: Night! |