Add --hash-style=gnu to LDFLAGS
Il giorno sab, 07/08/2010 alle 19.32 +0300, Markos Chandras ha scritto:
> > It seems like few of our fellow developers don't know how to track > down > packages that don't respect LDFLAGS. Adding -Wl,--hash-style=gnu is a > good way > to do that. I would like to see this linker flag enabled by default on > LDFLAGS > (or at least for the dev/ profiles for now). Do you agree? > It really depends on which glibc the profiles have been using: - >= 2.5 -> force --hash-style=gnu, reduced link time and smaller files; - < 2.5 -> force --hash-style=sysv, as above, but gnu would make lose the hash entirely. Somebody who works with uclibc should let us know which one of the two is actually supported by which uclibc. > -- Diego Elio Pettenò — “Flameeyes” http://blog.flameeyes.eu/ If you found a .asc file in this mail and know not what it is, it's a GnuPG digital signature: http://www.gnupg.org/ |
Add --hash-style=gnu to LDFLAGS
On Saturday 07 August 2010 19:48:35 Markos Chandras wrote:
> Is there another way to deal with this? Advertising it is enough IMHO |
Add --hash-style=gnu to LDFLAGS
On Saturday, August 07, 2010 12:32:31 Markos Chandras wrote:
> It seems like few of our fellow developers don't know how to track down > packages that don't respect LDFLAGS. Adding -Wl,--hash-style=gnu is a good > way to do that. I would like to see this linker flag enabled by default on > LDFLAGS (or at least for the dev/ profiles for now). Do you agree? it isnt a valid flag for everyone. it requires a new enough binutils (not too big of a deal), a new enough glibc (also not too big of a deal), and you cant be a mips target (glibc support is broken). support does exist in uClibc, but only in recent versions (this is an issue), and hasnt been widely tested for many targets. it also can cause binary compatibility issues for people who wish to compile a package on Gentoo but deploy elsewhere. not sure how much of a hassle this is for people. considering the only real benefit is that it automates a QA check, i dont think it's worth the hassle. newer Gentoo binutils already enables .gnu.hash support by default, so people get the speed increase transparently. > I would also like to know the procedures I should follow to make this > default on every profile as well. Do I need a council approval for that? not really -mike |
Add --hash-style=gnu to LDFLAGS
On Saturday, August 07, 2010 12:37:50 Alexis Ballier wrote:
> > It seems like few of our fellow developers don't know how to track down > > packages that don't respect LDFLAGS. Adding -Wl,--hash-style=gnu is a > > good way to do that. I would like to see this linker flag enabled by > > default on LDFLAGS (or at least for the dev/ profiles for now). Do you > > agree? > > no > > http://archives.gentoo.org/gentoo-dev/msg_27dfa5198d43c5ef1b4bfdc7d19e970f. > xml hmm, i thought i remembered answering this question before. figured i was just making things up in my head ;). -mike |
Add --hash-style=gnu to LDFLAGS
On Sat, Aug 07, 2010 at 05:32:42PM -0400, Mike Frysinger wrote:
> On Saturday, August 07, 2010 12:32:31 Markos Chandras wrote: > > It seems like few of our fellow developers don't know how to track down > > packages that don't respect LDFLAGS. Adding -Wl,--hash-style=gnu is a good > > way to do that. I would like to see this linker flag enabled by default on > > LDFLAGS (or at least for the dev/ profiles for now). Do you agree? > > it isnt a valid flag for everyone. it requires a new enough binutils (not too > big of a deal), a new enough glibc (also not too big of a deal), and you cant > be a mips target (glibc support is broken). support does exist in uClibc, but > only in recent versions (this is an issue), and hasnt been widely tested for > many targets. > We could at least enable that on x86/amd64 dev profiles (since only devs are supposed to use them and they are the most used arches among us) just to get some more feedback and fix the packages before they reach the end users. This will reduce the bug # and make sure that --as-needed is respected in a broader range of packages > > considering the only real benefit is that it automates a QA check, i dont > think it's worth the hassle. newer Gentoo binutils already enables .gnu.hash > support by default, so people get the speed increase transparently. Agreed -- Markos Chandras (hwoarang) Gentoo Linux Developer Web: http://hwoarang.silverarrow.org Key ID: 441AC410 Key FP: AAD0 8591 E3CD 445D 6411 3477 F7F7 1E8E 441A C410 |
Add --hash-style=gnu to LDFLAGS
On Sat, Aug 7, 2010 at 5:11 PM, Markos Chandras <hwoarang@gentoo.org> wrote:
> On Sat, Aug 07, 2010 at 05:32:42PM -0400, Mike Frysinger wrote: >> On Saturday, August 07, 2010 12:32:31 Markos Chandras wrote: >> > It seems like few of our fellow developers don't know how to track down >> > packages that don't respect LDFLAGS. Adding -Wl,--hash-style=gnu is a good >> > way to do that. I would like to see this linker flag enabled by default on >> > LDFLAGS (or at least for the dev/ profiles for now). Do you agree? >> >> it isnt a valid flag for everyone. *it requires a new enough binutils (not too >> big of a deal), a new enough glibc (also not too big of a deal), and you cant >> be a mips target (glibc support is broken). *support does exist in uClibc, but >> only in recent versions (this is an issue), and hasnt been widely tested for >> many targets. >> > We could at least enable that on x86/amd64 dev profiles (since only devs are > supposed to use them and they are the most used arches among us) just to get some more > feedback and fix the packages before they reach the end users. This will > reduce the bug # and make sure that --as-needed is respected in a broader > range of packages Why not just set some LDFLAGS that totally won't work (-Wl, taters) and then assume anything that actually compiles with those flags set does not respect LDFLAGS. Set that up on a tinderbox instead of making users do it. -A >> >> considering the only real benefit is that it automates a QA check, i dont >> think it's worth the hassle. *newer Gentoo binutils already enables .gnu.hash >> support by default, so people get the speed increase transparently. > Agreed > > > -- > Markos Chandras (hwoarang) > Gentoo Linux Developer > Web: http://hwoarang.silverarrow.org > Key ID: 441AC410 > Key FP: AAD0 8591 E3CD 445D 6411 3477 F7F7 1E8E 441A C410 > |
Add --hash-style=gnu to LDFLAGS
Il giorno dom, 08/08/2010 alle 03.11 +0300, Markos Chandras ha scritto:
> We could at least enable that on x86/amd64 dev profiles (since only > devs are > supposed to use them and they are the most used arches among us) just > to get some more > feedback and fix the packages before they reach the end users. This > will > reduce the bug # and make sure that --as-needed is respected in a > broader > range of packages I've set it out on the tinderbox, I can start reporting bugs from there... the problem is: _who_ is going to look at them? You can already deduce a subset of packages not respecting ldflags by taking the bugs that I report with forced --as-needed, take out those reported for LDFLAGS=-Wl,--as-needed as well: what remains is usually stuff that uses filter-ldflags or simply ignores the LDFLAGS variable. Not all of them, but some of them for a start. -- Diego Elio Pettenò — Flameeyes http://blog.flameeyes.eu/ |
Add --hash-style=gnu to LDFLAGS
On Sat, Aug 07, 2010 at 05:15:04PM -0700, Alec Warner wrote:
> > Why not just set some LDFLAGS that totally won't work (-Wl, taters) > and then assume anything that actually compiles with those flags set > does not respect LDFLAGS. Set that up on a tinderbox instead of > making users do it. > > -A Alec I was only talking about the *dev* profiles. As I said this will help *developers* to track this problem and fix it before they commit their packages to tree. Tinderbox will only catch packages that are already in the tree. The main goal here is to fix these packages before they reach it -- Markos Chandras (hwoarang) Gentoo Linux Developer Web: http://hwoarang.silverarrow.org Key ID: 441AC410 Key FP: AAD0 8591 E3CD 445D 6411 3477 F7F7 1E8E 441A C410 |
Add --hash-style=gnu to LDFLAGS
Il giorno sab, 07/08/2010 alle 17.15 -0700, Alec Warner ha scritto:
> > Why not just set some LDFLAGS that totally won't work (-Wl, taters) > and then assume anything that actually compiles with those flags set > does not respect LDFLAGS. Set that up on a tinderbox instead of > making users do it. Because such negative tests don't get far away: a single package failing will drop its whole deptree from merging. And we have much more important things to look for with a tinderbox than this, given the amount of (failure) builds you'd be expected to see with this approach. -- Diego Elio Pettenò — Flameeyes http://blog.flameeyes.eu/ |
Add --hash-style=gnu to LDFLAGS
On Saturday, August 07, 2010 20:11:42 Markos Chandras wrote:
> On Sat, Aug 07, 2010 at 05:32:42PM -0400, Mike Frysinger wrote: > > On Saturday, August 07, 2010 12:32:31 Markos Chandras wrote: > > > It seems like few of our fellow developers don't know how to track down > > > packages that don't respect LDFLAGS. Adding -Wl,--hash-style=gnu is a > > > good way to do that. I would like to see this linker flag enabled by > > > default on LDFLAGS (or at least for the dev/ profiles for now). Do you > > > agree? > > > > it isnt a valid flag for everyone. it requires a new enough binutils > > (not too big of a deal), a new enough glibc (also not too big of a > > deal), and you cant be a mips target (glibc support is broken). support > > does exist in uClibc, but only in recent versions (this is an issue), > > and hasnt been widely tested for many targets. > > We could at least enable that on x86/amd64 dev profiles (since only devs > are supposed to use them and they are the most used arches among us) just > to get some more feedback and fix the packages before they reach the end > users. This will reduce the bug # and make sure that --as-needed is > respected in a broader range of packages obviously you only mean linux x86/amd64 dev profiles. i dont have a strong opinion on that small subset in either direction. -mike |
| All times are GMT. The time now is 01:21 PM. |
VBulletin, Copyright ©2000 - 2013, Jelsoft Enterprises Ltd.
Content Relevant URLs by vBSEO ©2007, Crawlability, Inc.