mozilla :: #build

9 Oct 2017
04:59sajattackthere's something wrong with binutils
05:00sajattack ../../build/unix/gold/ld: internal error in override_version, at /build/binutils/src/binutils-gdb/gold/
05:00sajattackis that my binutils or the one bundled?
05:00sajattackit looks like the one bundled
05:31sajattackI resolved that issue by copying my ld to obj/build/unix/gold
07:17sajattackfor some reason my system is looking for gnu/stubs-soft.h when I have mfloat-abi=hard set
07:22sajattackresulting in a soft float
07:56sajattackHow can I force hard float?
10:42tedsajattack: what are you building for anyway?
10:42ted(just curious)
10:42sajattackraspberry pi
10:43sajattackI did a searchfox and found all the places softfp was defined
10:43sajattackshould be good now
10:44tedthere's --with-soft-float
10:45tedi think you can --with-soft-float=no
10:45tedthe rest of the options in that file might be of interest as well
10:45tedi think that's where we define all the arm arch-specific features
10:45tedi kinda wish we had official rpi builds, honestly
10:45sajattackwhere would I put --with-soft-float=no?
10:45tedin a mozconfig
10:48sajattackI wish we had official rpi builds too. Would've saved me a lot of trouble
10:54sajattackted: when I get my build done what do I copy to the pi? The whole dist folder?
10:54Fallen|ptotry mach package
10:54sajattackoh cool
10:55Fallen|ptoshould create a new subdir, on mac it is dist/ but obviously that is different for linux. I think maybe dist/firefox/ ?
11:22sajattack--with-soft-float=no translates to -mno-soft-float which isn't supported on armv7a
11:22sajattackI think it's a aarch64 thing
11:23sajattack^ ted
11:24sajattackbut I found MOZ_FLOAT_ABI
11:25sajattackand a bunch of places in webrtc that force softfp for some reason
14:54dmajorted: ping; I installed VS2017 but my build isn't fully picking it up; my config.status is a frankenstein
15:04teddmajor: exciting
15:04teddmajor: you're using start-shell.bat?
15:05dmajorted: yes, and --with-visual-studio-version=2017 for good measure
15:05tedyou clobbered, and it still came out weird?
15:05dmajoryeah it was a fresh objdir
15:07tjrI have a question/problem with sccache... it's giving bad output
15:07tjrspecifically, it's not picking up a change I made in a header file here:
15:07tjrIf you look at the logs, you'll find 'warning: "HAVE_PTHREAD_H" redefined' - I resolved these by editing HAVE_PTHREAD_H in two .h files. However, these changes are not being picked up by sccache
15:07tjrI verified by using a one click loaner and ran the individual compile statement with sccache (got warning) and without (no warning)
15:09tedtjr: so sccache runs the preprocessor and uses the output as part of the cache key
15:09tedif you get the same preprocessor output with your change, and you didn't change the compiler commandline, sccache is still going to get a cache hit
15:09tedand unfortunately if it's cached you'll get the same compiler output
15:10tjrAhha. That explains it. I can hack around it to do a successfull try run, but does this mean in my 'official' patch I need to introduce a superficial change in the actual code? Is there a standard technique for this?
15:11tedthere is not, AFAIK
15:11tedsccache cache entries only persist for ~2 weeks, fwiw
15:11tedand if anyone makes a change in the actual source in the interim it'll change things
15:11tedyou can certainly make a whitespace or formatting change or something to make it so it's a cache miss
15:12tjrI'm not expecting that to happen so I'll try and see what I can do :)
15:12tedok :)
15:17teddmajor: OK, so,
15:17tedis where it does the VC detection
15:22dmajorted: running those steps by hand seems to give me the right results
15:22teddmajor: OK, so what looks wrong in your config.status?
15:23dmajorted: link is right but cl is not
15:24teddmajor: bleh
15:24teddmajor: what's PATH set to in config.status?
15:24tedi think _CC / _CXX come from old-configure
15:28dmajorHm, c:\\Program Files (x86)\\Microsoft Visual Studio 14.0\\VC\\BIN\\amd64 is above C:/Program Files (x86)/Microsoft Visual Studio/2017/Community/VC/Tools/MSVC/14.11.25503/bin/HostX64/x64
15:39dmajorted: oh no
15:40dmajorted: this is what I think happened: I muscle-memoried my way to start-shell-vs2015, then I realized my mistake, exited, and launched start-shell - but! I still had MOZ_MSVCVERSION=14 from my earlier shell
15:40dmajorI think it's working now :-/
15:44dmajoryep, configure does all the right things, when I use a fresh cmd
15:45tedwould certainly be good if we had code to catch that
15:45dmajorMaybe I should make a start-shell-2017 that unsets everything that 2015 sets
15:48dmajorseems to work!
15:52dmajorI suppose the more proper thing would be for start-shell-2015 to do a SETLOCAL
15:53dmajorRyanVM: does such a file still exist in mozillabuild 3.0? ^
15:56dmajorgreat, no need to patch it then :-D
15:58tedgenerally the configure detection of MSVC is good but not perfect
15:58tedwould be great if we had an initial bit that detected an MSVC install, and then everything else fed from that instead of using PATH
15:58dmajorted: now it looks like my 2017 linker is getting picked up for my 2015 builds. is that expected?
16:12dmajorted: do you have both compilers installed? can you check what you get for link.exe?
16:12tedi do have both installed
16:12tedlink.exe in a build from a start-shell-2015.bat shell?
16:14dmajor 0:24.28 checking for link... 'C:/PROGRA~2/MICROS~1.0/VC/bin/amd64/link.exe'
16:14dmajor 1:00.71 js\src> checking for link... 'C:/PROGRA~2/MICROS~2/2017/COMMUN~1/VC/Tools/MSVC/1411~1.255/bin/HostX64/x64/link.exe'
16:16RyanVMoh my
16:16RyanVMI'm sure that'll work out swimmingly
16:17dmajor^ the above is if I say --with-visual-studio-version=2015
16:17dmajorIf I omit --with-visual-studio-version, I get 2017 for both (which I guess kinda makes sense)
16:20tjrted: Is the patch in sufficient? I couldn't quite glean from your link what I should be doing if not disabling the check step...
16:20firebotBug 1395047 NEW, --enable-tests for MinGW build is broken because it thinks it's linux
16:20dmajorIf I use a start-shell (no version), and say --with-vs=2015, I also get a 2015 linker on the top-level configure and 2017 in the js/src
16:20teddmajor: ugh
16:21tedtjr: yeah, that should be fine
16:21tedthat's what we did for cross-mac
16:26dmajorted: is this another js needs to re-define a top-level somethingorother mumble mumble handwave?
18:39* dmajor pokes ted
18:41teddmajor: sorry, got tied up in something else
18:52teddmajor: i think we just pass down whatever's in CC/CXX to js' configure
19:00teddmajor: OK yeah, i can reproduce that here
19:01teddmajor: ohh
19:02tedi bet --with-visual-studio-version is not a js_option
19:02tedso js/src/configure doesn't respect it
19:02* dmajor price_is_right_fail.wav
19:06teddmajor: file it?
19:07* Aryx wonders if americans know this sound:
19:16dmajorted: confirmed that js_option works, thanks! filed 1407017 with patch
19:23dmajorted: out of curiosity, why isn't every option a js_option?
19:23tedbecause standalone js builds don't care about *everything*
19:23dmajoroh, right, standalone builds
19:26sajattackI'm getting undefined references to a bunch of webrtc stuff
19:27sajattack/home/paul/ff-src/firefox-57.0b4/media/webrtc/trunk/webrtc/common_audio/ undefined reference to `webrtc::FIRFilterNEON::FIRFilterNEON(float const*, unsigned int, unsigned int)'
19:27sajattackand on and on
19:28sajattackit's all audio related stuff
20:04tedsajattack: that code apparently thinks you're using NEON, but it's not building the NEON code?
20:16sajattackted: sorry my computer crashed. Yeah I am using neon. I undid some changes and clobbered to see if that would help, still waiting on it to rebuild
20:17sajattackI was having issues with aom_convolve8 earlier and removed it
20:17sajattackI think that might have something to do with it
20:18tedthat's what selects that NEON code
20:19tedwe only turn that on for android apparently
20:29sajattackhow do mozillians groups work?
20:29sajattackLike, is there somewhere people in the same group chat?
20:29sajattacksorry for offtopic
20:57sajattackdamn I'm still having issues
21:02sajattackdid I turn on something that should only be turned on for android by accident?
21:11sajattackoh I think I fixed it
21:12mshalwhat was the problem?
21:12sajattackby changing my mozbuild to force neon rather than optional neon
21:13sajattackmshal: I was having errors with libxul not being able to find neon audio stuffs
21:16sajattackyay everything compiles
21:17mshaldid you find out why it was building with WEBRTC_HAS_NEON but not building the neon code?
21:18sajattackmshal: like I said, I just switched from optional neon to forced neon
21:26sajattackrip it segfaults
21:29sajattackhow do I debug
21:40sajattack_I've never really used gdb
21:40sajattack_is there a guide somewhere?
21:44sajattack_gdb ./firefox doesn't print anything
21:44sajattack_would a strace be helpful?
21:45mshalpoking around for &#39;gdb tutorial&#39; might be good. Personally I might just do &#39;gdb <program name>&#39; and then inside gdb do &#39;r&#39; to run it, and when it crashes do &#39;bt&#39; to get a backtrace
21:46mshalI don&#39;t think strace is likely to help initially, unless you find out there&#39;s a problem in a system call
21:49mshalmight need to build with --enable-debug?
21:49mshalalso forgot to mention, but for multithreaded programs &#39;thread apply all bt&#39; instead of &#39;bt&#39; would give the backtrace for all threads
22:08MelchiorGasparI saw that FF v56.0.1 is out.. but when is ESR v52.4.1 out? the candiate files are there... :/
22:17RyanVMMelchiorGaspar: in the next day or two
22:17RyanVMbut it&#39;s only interesting if you&#39;re on OSX 10.13
22:17RyanVM(all 3 fixes that went into it were to address issues with 10.13)
22:18gandalfgps|away: can you give me any ETA for bug 1394891? It&#39;s a blocker for a release and I&#39;d like to plan around it
22:18firebot ASSIGNED, Add a temporary whitelist of files localized via L20n to control the exposure
22:18glandiumno need for --enable-debug to debug
22:19MelchiorGasparoh.. lol >_> ok thx.. ahh.. no fun...
22:19glandiumsajattack_, mshal: ^
22:19glandiumsajattack: no need for --enable-debug to debug
22:19sajattackthen why wasn&#39;t it printing anything helpful?
22:20glandiumsajattack: you haven&#39;t said much about what you tried
22:20sajattackgdb ./firefox r bt
22:20glandiumsajattack: and what was the output?
22:21glandiumsajattack: you&#39;re running firefox from where it&#39;s installed with make install, aren&#39;t you?
22:22glandiumthat strips the executables, so you lose the debug symbols
22:22sajattackit&#39;s not installed with make install
22:22sajattackit&#39;s make package&#39;d then copied over
22:22glandiumsame thing
22:22sajattackso what do I do?
22:22glandiummach run
22:23sajattackbut then I have to copy mach and everything
22:23glandiumthat wouldn&#39;t work
22:23glandiumyou need to run an unstripped firefox
22:23glandiumso, you might as well run the one in your build directory
22:23glandiumoh, you&#39;re cross-compiling
22:24glandiumso what you want is --disable-install-strip
22:24glandiumyes ; if you don&#39;t want to rebuild just for that, you can do make package PKG_SKIP_STRIP=1
22:26sajattackI&#39;ve still got the build with --enable-debug going
22:26glandiumthat&#39;s likely a waste of time
22:27sajattackoh well
22:27sajattackit&#39;s almost done
22:36sajattackglandium: ^
22:37glandiumsajattack: make -C obj-whatever package PKG_SKIP_STRIP=1
23:30sajattackgdb chews ram like chrome
23:33sajattacknot sure what it&#39;s doing
23:35glandiumsajattack: reading debug symbols
23:35sajattacka lot of them apparently
23:35glandiumsajattack: look at the size of
23:58sajattackhow long is this supposed to take?
23:59sajattackit&#39;s been going for like half an hour
10 Oct 2017
No messages
Last message: 7 days and 23 hours ago