Summary: | www-client/firefox-7.0.1-r1 fails to build (with USE=pgo?) | ||
---|---|---|---|
Product: | Gentoo Linux | Reporter: | JMW <jessemichaelwilson> |
Component: | Current packages | Assignee: | Mozilla Gentoo Team <mozilla> |
Status: | RESOLVED FIXED | ||
Severity: | major | CC: | ainsaar, bircoph, boxcars, d.rammelt, duncanphilipnorman, jessemichaelwilson, johnnybit, phmagic, Tanktalus |
Priority: | Normal | Keywords: | PATCH |
Version: | unspecified | ||
Hardware: | AMD64 | ||
OS: | Linux | ||
Whiteboard: | |||
Package list: | Runtime testing required: | --- | |
Attachments: |
ebuild environment file
build.log.gz Output of dmesg >> ~/dmesg.log Latest build.log.gz. firefox-48.0-pgo.patch firefox-48.0.ebuild.patch |
Description
JMW
2011-10-26 17:47:42 UTC
URL for build.log (8MB): http://www.2shared.com/file/tGaCALqw/build.html Created attachment 290917 [details]
ebuild environment file
Environment file.
Update: It works if I emerge it with pgo disabled, so this is definitely a pgo issue. Do not use third party hosting sites for bug reports. Please post your build log as an attachment. You can use gzip to compress it first if necessary. Created attachment 290951 [details]
build.log.gz
Sorry, when I tried to post it originally B.G.O complained about the 1000kb limit, and said I could either compress it OR host it on a different server. I wasn't aware that 8000kb would compress to under 1000kb, so I assumed the proper thing to do would be to have it hosted elsewhere. I don't actually see an error message in that build log. It just stops after zipping up a bunch of files. Are you sure you attached the right one? 100%. I was equally confused. Okay then. Maybe it will make sense to the Mozilla team. :) Just let me know if you need me to try something out and post the results. I'm anxious to get pgo enabled firefox working. (In reply to comment #10) > Just let me know if you need me to try something out and post the results. I'm > anxious to get pgo enabled firefox working. Unless you can show where the error is, we are not gonna be able to do anything for you. Check dmesg for failure message related to build. Something has to be segfaulting somehwere. Casually browsing dmesg output, I did see one segfault, relating to libflashplayer.so. I suppose it's possible firefox includes flash performance in its PGO, so this might be it, although it's weird that I'm not seeing (or maybe just don't notice) flash segfaulting normally. More info, when I try the PGO build from my xsession, it fails because no usable screens are found or something. So this is me emerging PGO firefox from the console and failing without useful error message. My xsessions (multiple) are started by Qingy if that helps anything. I'm attaching the dmesg.log resulting from dmesg >> ~/dmesg.log. While you look at that, I'll be trying to rebuild libflashplayer.so and then reemerge firefox. I will be changing locations tomorrow and may not have my computer setup for a bit, so hopefully we will get some good info from this. If not, I'll continue to help solve this bug as soon as I'm able. Created attachment 291265 [details]
Output of dmesg >> ~/dmesg.log
All I saw was the libflashplayer error, but that was from a very cursory glance. You may well see more.
Upgrading to the latest unstable flashplayer (which either doesn't require, or else incorrectly doesn't depend on nspluginwrapper) I was able to get a more useful error message. I'm attaching the latest build.log.gz. After this, I will try emerging firefox with PGO, but without flashplayer on my system at all. I'm starting to suspect we might have this bug licked pretty soon... Created attachment 291321 [details]
Latest build.log.gz.
This is the build.log.gz from emerging firefox with USE=pgo after having emerged the latest (unstable) flashplayer without nspluginwrapper on my system. This is with 64bit and -32bit useflags set on flashplayer.
That last error was unrelated to the original issue. It had to do with a service running on the same port as the EasyServer python function that PGO makes use of. I've killed the server, and I'm now attempting an emerge of Firefox WITH the latest flashplayer. It worked! A possible resolution for this could be to make the latest Firefox with PGO enabled depend on the latest adobe-flash (11.0.1.152), although I suspect this will not be the way to go. Probably just a simple elog message, something to the effect of: "Adobe-Flash versions <11.0.1.152 may make PGO enabled Firefox fail to build. If you think this is the case, confirm by checking dmesg output for an error involving libflashplayer.so and then emerge >=adobe-flash-11.0.1.152". I'm not filing a separate bug report on the issue regarding conflicting services running because I have a lot of things to do right now, but if anyone cares, an appropriate ewarn could be placed in the ebuild regarding any services running on port 8888 preventing the PGO build from completing. If you need to confirm that it is port 8888 just look in the source tree under mozilla-release/obj-x86_64-unknown-linux-gnu/_profile/pgo/profileserver.py. Not far down in the file it will tell you which port it uses. You could even make the ebuild check that file for the actual port number and warn appropriately, or if you wanted to get sophisticated, just have the ebuild make the appropriate sed's and allow the user to pass a custom port number to the PGO build in an environment variable. I'm marking these two bugs "worksforme" in my own personal bugzilla, so I hope other users will be able to benefit from this bug report somehow. I see this with firefox 10 with pgo and had freenet running. Thanks for the port 8888 tip as I knew it was freenet on my system stopping that got me past that. *** Bug 402265 has been marked as a duplicate of this bug. *** *** Bug 407043 has been marked as a duplicate of this bug. *** *** Bug 409253 has been marked as a duplicate of this bug. *** I am facing a very similar problem except that I am building with -pgo. By the very similar problem I mean that the build stops with: run-mozilla.sh: Cannot execute /tmp/portage/www-client/firefox-10.0.3/work/mozilla-esr10/obj-x86_64-unknown-linux-gnu/browser/installer/../../dist/bin/shlibsign. (In reply to comment #22) > I am facing a very similar problem except that I am building with -pgo. > > By the very similar problem I mean that the build stops with: > > run-mozilla.sh: Cannot execute > /tmp/portage/www-client/firefox-10.0.3/work/mozilla-esr10/obj-x86_64-unknown- > linux-gnu/browser/installer/../../dist/bin/shlibsign. Same here, building with -pgo. In which way is this bug related to bug 409331 ? (In reply to comment #23) > (In reply to comment #22) > > I am facing a very similar problem except that I am building with -pgo. > > > > Same here, building with -pgo. > > In which way is this bug related to bug 409331 ? Same here, building with -pgo. Getting the same error as bug 409253... not sure why it's listed as a dup of this. PGO useflag has been masked until such a time we can properly support it in gentoo. Soon as we can we will unmask and support pgo builds for all users. At this time even upstream is having problems supporting pgo builds in Linux. I will continue to track there upstream to determine when best we can come back and review all pgo build failures. Please ignore resolution we no longer have an option for later. *** Bug 499650 has been marked as a duplicate of this bug. *** @mozilla team, I understand that pgo is PITA and you can't fully support it right now. But please, at least accept provided patches, at least if they don't affect non-pgo builds: this way you will keep users willing to use pgo happy and will not hurt others. This will also make pgo more tested and may result in its unmasking in future. I ported patches provided by Daniel Rammelt in bug 499650 for firefox-48, see below. Works fine for me on ~amd64 (both build and run-time). Created attachment 442808 [details, diff]
firefox-48.0-pgo.patch
Adds pgo run target.
Created attachment 442810 [details, diff]
firefox-48.0.ebuild.patch
(In reply to Andrew Savchenko from comment #27) > @mozilla team, I understand that pgo is PITA and you can't fully support it > right now. But please, at least accept provided patches, at least if they > don't affect non-pgo builds: this way you will keep users willing to use pgo > happy and will not hurt others. This will also make pgo more tested and may > result in its unmasking in future. > > I ported patches provided by Daniel Rammelt in bug 499650 for firefox-48, > see below. Works fine for me on ~amd64 (both build and run-time). You know, since you're a dev, you're welcome to join the mozilla team and manage the PGO patches directly.. :) (In reply to Ian Stakenvicius from comment #30) > You know, since you're a dev, you're welcome to join the mozilla team and > manage the PGO patches directly.. :) Thank you for the invitation, I'm joining you to maintain the PGO stuff :) I'm going to apply mentioned patches and unmask pgo on amd64 starting from 48.0. I have troubles on ~x86 (segfaults) and I can't test other arches, so it will be kept masked there. *** Bug 480258 has been marked as a duplicate of this bug. *** Nice work! However, I think my patch from bug 480258 comment 18 would be easier to maintain in the future. Upstream requires tests to be enabled for PGO builds [1]. I'll see what are implications of this. [1] https://developer.mozilla.org/en-US/docs/Mozilla/Developer_guide/Build_Instructions/Building_with_Profile-Guided_Optimization#Getting_Started @Andrew Savchenko Please, see bug 590658. Normal build with pax is fixed now, but pgo build is failing as described there. |