View Issue Details
ID | Project | Category | View Status | Date Submitted | Last Update |
---|---|---|---|---|---|
0028448 | Lazarus | Compiler | public | 2015-07-31 12:40 | 2020-10-15 14:09 |
Reporter | Graham Inggs | Assigned To | Juha Manninen | ||
Priority | normal | Severity | minor | Reproducibility | always |
Status | resolved | Resolution | fixed | ||
Platform | ARM | OS | Debian | ||
Product Version | 1.4 | ||||
Summary | 0028448: lazbuild appears to stall when building on armel and armhf | ||||
Description | Since lazarus 1.4.0 appeared in the Debian unstable distribution, builds on armel and armhf have been failing, appearing to stall. Full build logs can be viewed [1][2][3] by clicking on the Build-Attempted links for a particular architecture. [1] https://buildd.debian.org/status/package.php?p=doublecmd&suite=unstable [2] https://buildd.debian.org/status/package.php?p=easymp3gain&suite=unstable [3] https://buildd.debian.org/status/package.php?p=cqrlog&suite=unstable | ||||
Steps To Reproduce | I chose a simpler project [1] and attempted to build it on real ARM hardware (NVIDIA Jetson TK1). Instead of building the Debian package, I simply ran: lazbuild source/ddrescueview.lpi --bm="GNU/Linux Release" The build was successful with lazarus 1.2.4 and 1.2.6, but appeared to stall with lazarus 1.4.0 and 1.4.2. In all cases, fpc 2.6.4 as it is in Debian was used. I attach build logs of the 1.2.6, 1.4.0 and 1.4.2 builds. I noticed that the failed build logs were 4096 bytes in size and, as can be seen in the logs from the Debian buildds, the last line of the log is truncated, e.g. TExternalTool.DoExecute Title="Compile Project, Mode: GNU/Linux Release, Target: ddrescueview" Process.CurrentDirectory="/srv/ddrescueview/source/" Executable="/usr/bin/fpc" Params: -B -Tlinux -MObjFPC -Schi -CX -O3 -Xs -XX -k-z -krelro -k--as-needed -l -vewnhibq -Fl/opt/gnome/lib -Fu/usr/lib/lazarus/1.4.2/lcl/units/arm-linux/gtk2 -Fu/usr/lib/lazar After hitting ctrl-c, I noticed that the build had actually completed successfully and there was a working 'ddrescueview' executable. [1] http://ddrescueview.sourceforge.net/ | ||||
Tags | No tags attached. | ||||
Fixed in Revision | r64005 | ||||
LazTarget | - | ||||
Widgetset | |||||
Attached Files |
|
related to | 0036318 | resolved | Juha Manninen | lazbuild crashes when compiling packages |
|
|
|
Manually running the fpc command line that lazbuild runs (see below), seems fine, and the output appears on stdout as expected. /usr/bin/fpc -B -Tlinux -MObjFPC -Schi -CX -O3 -Xs -XX -k-z -krelro -k--as-needed -l -vewnhibq -Fl/opt/gnome/lib -Fu/usr/lib/lazarus/1.4.2/lcl/units/arm-linux/gtk2 -Fu/usr/lib/lazarus/1.4.2/lcl/units/arm-linux -Fu/usr/lib/lazarus/1.4.2/components/lazutils/lib/arm-linux -Fu/usr/lib/lazarus/1.4.2/packager/units/arm-linux -Fu/srv/ddrescueview/source/ -dLCL -dLCLgtk2 ddrescueview.lpr |
|
On ARM, the following call from around line 510 in ide/exttools.pas never returns: Thread.Synchronize(Thread,@NotifyHandlerNewOutput); Any ideas where else I can look? |
|
--max-process-count=0 bug in lazbuild? Automatic detection of CPU count sometimes fails. try to run with lazbuild --max-process-count=1 |
|
Thanks, but that option doesn't seem to be available in lazarus 1.4.2: Invalid option at position 1: "max-process-count" |
|
I tried merging r48422 [1] (which added --max-process-count) and r48447 [2] (which I believe fixed the issue Anton was referring to) into lazarus 1.4.2 and still experienced the problem, even when setting --max-process-count=1. I also tried the lazarus 1.5-49593-20150805 snapshot [3] and the problem is still present. [1] http://svn.freepascal.org/cgi-bin/viewvc.cgi?view=revision&root=lazarus&revision=48422 [2] http://svn.freepascal.org/cgi-bin/viewvc.cgi?view=revision&root=lazarus&revision=48447 [3] ftp://ftp.freepascal.org/pub/lazarus/snapshots/lazarus-1.5-49593-20150805-src.zip |
|
Does the linker on that system suffer from https://sourceware.org/bugzilla/show_bug.cgi?id=18344 ? You should be able to verify that by compiling any program with fpc and seeing whether there are a bunch of stray empty lines. If so and if that output gets written to stderr rather than standard out, what possibly happens is that stderr gets full and stalls the child process when it tries to write more data to it. FPC (not Lazarus) revision 29303 contains a work around for child processes launched via the tprocess class getting stuck if they write a lot to stderr and little or nothing to stdout. Even if the linker does not suffer from the problem mentioned at the start of this comment, it could still be useful to try. |
|
> Does the linker on that system suffer from > https://sourceware.org/bugzilla/show_bug.cgi?id=18344 [^] ? > You should be able to verify that by compiling any program with fpc and > seeing whether there are a bunch of stray empty lines. Yes, it does suffer from that problem (I didn't even know it was a problem!). You can even see it in, for example, the i386 build of doublecmd from one of the links in my description, although it doesn't stall on that architecture. > FPC (not Lazarus) revision 29303 contains a work around for child processes > launched via the tprocess class getting stuck if they write a lot to stderr > and little or nothing to stdout. I'm going to try anyway, but do you think it would be sufficient just to merge in that commit? So far, I have been unable to build the fixes branch or the snapshot of fpc on this system. I can, however, rebuild the Debian fpc package with patches. |
|
> I'm going to try anyway, but do you think it would be sufficient just to merge in that commit? Yes, it should be independent of everything else. > So far, I have been unable to build the fixes branch or the snapshot of fpc on this system. What is the problem? |
|
> Yes, it should be independent of everything else. I ended up merging r25363, r28972 and r29303, rebuilding fpc and then rebuilding lazarus. Unfortunately lazbuild still appears to hang, and needs Ctrl-C. >> So far, I have been unable to build the fixes branch or the snapshot of fpc on this system. > What is the problem? There are many patches against the Debian fpc package [1], and I haven't been able to go through them all to determine which ones are still needed and get them to apply cleanly, so I couldn't easily drop in an updated fpc source tarball and build a new Debian package, like I could do with lazarus. I also tried 'NOGDB=1 make all' in the daily source snapshot of the fixes tree (fpcbuild.zip), but got the following error: flt_pack.inc(119,5) Error: Syntax error while parsing a conditional compiling expression system.inc(1821,8) Warning: Implicit string type conversion from "RawByteString" to "UnicodeString" system.pp(375) Fatal: There were 1 errors compiling module, stopping Unfortunately, I will not be able to do any further testing on this for the next three weeks or so. [1] https://anonscm.debian.org/cgit/pkg-pascal/fpc.git/tree/debian/patches |
|
The problem seems to be related to the '-z relro' linker option and ARM. If we build Lazarus with the Debian default hardening flags, which include '-z relro', then lazbuild appears to stall. If we rebuild Lazarus and filter out '-z relro' from the linker flags, lazbuild works as expected. |
|
Could you try with trunk r31451 merged? |
|
> Could you try with trunk r31451 merged? I was able to apply r31451 to fpc 2.6.4 and that worked, thanks! I have tested on what Debian call 'armhf', but do you think this should work for 'armel' as well? Unfortunately, I don't have any suitable hardware to test on. BTW, I still get the flt_pack.inc and system.inc errors mentioned above when tryng to build fpc 3.0.0 rc1. Should I open a separate bug for that? |
|
> I was able to apply r31451 to fpc 2.6.4 and that worked, thanks! Thanks, I'll ask to merge it into 3.0 as well then. > I have tested on what Debian call 'armhf', but do you think this should work for 'armel' as well? Yes. > BTW, I still get the flt_pack.inc and system.inc errors mentioned above when tryng to build fpc 3.0.0 rc1. The system.inc message is a warning, not an error, and is benign. I can't imagine where the flt_pack.inc error comes from, other than a bug in the Debian 2.6.4 ARMHF compiler binary (which uses patches that have been backported by Debian developers, and which are not supported by the FPC developers with FPC 2.6.4). The code in flt_pack.inc is used on all platforms and builds fine with FPC 2.6.4 everywhere else, as far as I know. |
|
I've merged r31451 to fixes-3_0. Can this bug be closed now? Or is it something unrelated? |
|
That's a good question. The title of the bug report mentions ARMEL and ARMHF, but r31451 only affects ARMHF. |
|
I just rebuild winff with fpc 3.0.0+dfsg-2 and lazarus 1.6~rc2+dfsg-1 and the issue doesn't manifest itself. So I would say this bug can be closed. |
|
Apparently fixed. Please test. |
Date Modified | Username | Field | Change |
---|---|---|---|
2015-07-31 12:40 | Graham Inggs | New Issue | |
2015-07-31 12:40 | Graham Inggs | File Added: ddrescueview-build-logs.tar.gz | |
2015-07-31 17:15 | Graham Inggs | Note Added: 0085118 | |
2015-08-04 11:38 | Graham Inggs | Note Added: 0085193 | |
2015-08-04 13:59 | Anton Kavalenka | Note Added: 0085197 | |
2015-08-04 14:00 | Anton Kavalenka | Note Edited: 0085197 | View Revisions |
2015-08-04 21:49 | Graham Inggs | Note Added: 0085207 | |
2015-08-06 00:51 | Graham Inggs | Note Added: 0085221 | |
2015-08-06 14:31 | Jonas Maebe | Note Added: 0085231 | |
2015-08-06 17:13 | Graham Inggs | Note Added: 0085237 | |
2015-08-06 17:57 | Jonas Maebe | Note Added: 0085238 | |
2015-08-07 14:10 | Graham Inggs | Note Added: 0085251 | |
2015-09-03 13:08 | Graham Inggs | Note Added: 0085686 | |
2015-09-13 00:21 | Jonas Maebe | Note Added: 0085868 | |
2015-09-15 13:37 | Graham Inggs | Note Added: 0085939 | |
2015-09-15 13:59 | Jonas Maebe | Note Added: 0085940 | |
2015-09-30 21:43 | Joost van der Sluis | Note Added: 0086193 | |
2015-10-01 07:53 | Jonas Maebe | Note Added: 0086194 | |
2016-02-27 21:52 | Paul Gevers | Note Added: 0090417 | |
2017-08-27 00:28 | Juha Manninen | Relationship added | related to 0031989 |
2017-08-28 20:06 | Juha Manninen | Relationship deleted | related to 0031989 |
2019-11-16 07:42 | Juha Manninen | Relationship added | related to 0036318 |
2020-10-15 14:09 | Juha Manninen | Assigned To | => Juha Manninen |
2020-10-15 14:09 | Juha Manninen | Status | new => resolved |
2020-10-15 14:09 | Juha Manninen | Resolution | open => fixed |
2020-10-15 14:09 | Juha Manninen | Fixed in Revision | => r64005 |
2020-10-15 14:09 | Juha Manninen | LazTarget | => - |
2020-10-15 14:09 | Juha Manninen | Note Added: 0126330 |