On 2017-11-09 10:38, Philippe Gerum wrote: > On 10/19/2017 06:55 PM, Henning Schild wrote: >> On Thu, 19 Oct 2017 12:03:33 +0200 >> Philippe Gerum <r...@xenomai.org> wrote: >> >>> On 10/13/2017 01:36 PM, Norbert Lange wrote: >>>> I works, because collect2 (the "compiler driver", which documents >>>> the -r option) now knows about the -r flag, using -Wl,-[U]r would >>>> mean the flags aren't even looked at. >>>> It can then disable functionality that is not supposed to be done at >>>> this point (see gcc/collect2.c, search for 'early_exit'). It might >>>> actually be in conflict with the -Ur option from the comments, but I >>>> don't know what this is trying to solve (the constructor tables >>>> would then be built in the second step?) >>>> >>>> I would prefer the compiler driver knowing about the flags and >>>> correctly dealing with them, instead of replicating this logic in >>>> scripts (and keeping it up to date with this internal logic). >>> >>> Ack. >>> >>>> Might be that your wrapper script is doing something similarly by >>>> filtering out the arguments for the first step, which might be the >>>> same functionally that is skipped when adding '-r' >>>> >>>> I dont know what the -Ur flag is trying to solve, and the first >>>> patch adding -no-pie is way less likely to change something (given >>>> that I would not have compiled with -pie before). >>>> The new patch would however leave more of the magic to the toolchain >>>> and seems more "correct" to me, for whatever that's worth. But it >>>> might have some subtle differences to before, got some testcase >>>> where lacking the -Ur flag made a difference? >>> >>> I don't know the original intent about passing -Ur to the linker, >>> except maybe to stick to the ld manpage which states that the last >>> partial link command should be given such option. >>> >>> However my understanding is that we don't need it in wrap-link.sh, >>> since the script eventually completes the link stage to produce a >>> fully resolved executable. That implies collecting the ctors/dtors and >>> resolving all references from the partially linked object file anyway. >>> >>> Some testing only passing -r here seems to confirm this assumption >>> with C++ apps; C++ users may want to check this too: >> >> I just told some C++ power users to give that a try. I am not sure how >> fast the feedback will arrive, i hope in the next two weeks. >> >> Henning >> >>> >>> diff --git a/scripts/wrap-link.sh b/scripts/wrap-link.sh >>> index aab7899..fa83bd5 100755 >>> --- a/scripts/wrap-link.sh >>> +++ b/scripts/wrap-link.sh >>> @@ -204,7 +204,7 @@ done >>> >>> if $stage2; then >>> $verbose && set -x >>> - $dryrun $cc -o "$output.tmp" -Wl,-Ur -nostdlib $stage1_args >>> + $dryrun $cc -o "$output.tmp" -r -nostdlib $stage1_args >>> $dryrun $cc -o "$output" "$output.tmp" $stage2_args >>> $dryrun rm -f $output.tmp >>> else >>> > > Any news from the Xenomai/C++ folks regarding this change? I plan to > merge it next week if nobody complains. >
(Henning is on vacation) What I heard from our folks who could test this extensively is that they are unfortunately busy right now. So, if you want to merge, you should probably not wait for this feedback. Jan -- Siemens AG, Corporate Technology, CT RDA ITP SES-DE Corporate Competence Center Embedded Linux _______________________________________________ Xenomai mailing list Xenomai@xenomai.org https://xenomai.org/mailman/listinfo/xenomai
