www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.learn - Huge increase in UT compile time

reply Dhananjay <ds ds.com> writes:
Hello,

I am upgrading to DMD 2.076.1 from DMD 2.069.2 (similar results 
on 2.075.1), and seeing a huge increase in unittest compilation 
time when the -deps parameter is also passed to dmd. This is on 
both OSX and linux. What can be the cause of this?

Sample program:

     import std.stdio: writeln;

     unittest { writeln("TestUT"); }

     version (unittest) {}
     else
     {
         void main()
         {
             writeln("TestMain");
         }
     }

Observations:

Command: time dmd -deps=test.dep -c -o- test.d -de -w -m64 -color 
-g -debug -gs -unittest -main
Linux runtime: user 0m28.192s    << Note the increase
OSX runtime: user 0m48.508s     <<
Linux 2.069.1 runtime: user 0m0.009s

Command: time dmd -c -o- test.d -de -w -m64 -color -g -debug -gs 
-unittest -main
Linux runtime: user 0m0.064s
OSX runtime: user 0m0.090s
Linux 2.069.1 runtime: user 0m0.005s

Command: time dmd -deps=test.dep -c -o- test.d -de -w -m64 -color 
-g -debug -gs
Linux runtime: user 0m0.584s
OSX runtime: user 0m0.882s
Linux 2.069.1 runtime: user 0m0.007s

Command: time dmd -c -o- test.d -de -w -m64 -color -g -debug -gs
Linux runtime: user 0m0.048s
OSX runtime: user 0m0.074s
Linux 2.069.1 runtime: user 0m0.010s

Environment:
OSX: El Capitan 10.11.6
Linux with DMD 2.076.1: Gentoo 4.9.34
Linux with DMD 2.069.1: Centos 7
Oct 10
next sibling parent Dhananjay <ds ds.com> writes:
Another observation:
wc -l test.dep (for "-deps=test.dep ... -unittest -main" command):
Linux 2.076.1: 41389
Linux 2.069.1: 144

Making small changes to 2.069.1 runtimes (I had run with a 
different program by mistake. The measurement is still of the 
same order)

On Wednesday, 11 October 2017 at 06:25:19 UTC, Dhananjay wrote:
 Hello,

 I am upgrading to DMD 2.076.1 from DMD 2.069.2 (similar results 
 on 2.075.1), and seeing a huge increase in unittest compilation 
 time when the -deps parameter is also passed to dmd. This is on 
 both OSX and linux. What can be the cause of this?

 Sample program:

     import std.stdio: writeln;

     unittest { writeln("TestUT"); }

     version (unittest) {}
     else
     {
         void main()
         {
             writeln("TestMain");
         }
     }

 Observations:

 Command: time dmd -deps=test.dep -c -o- test.d -de -w -m64 
 -color -g -debug -gs -unittest -main
 Linux runtime: user 0m28.192s    << Note the increase
 OSX runtime: user 0m48.508s     <<
Linux 2.069.1 runtime: user 0m0.045s <edited>
 Command: time dmd -c -o- test.d -de -w -m64 -color -g -debug 
 -gs -unittest -main
 Linux runtime: user 0m0.064s
 OSX runtime: user 0m0.090s
Linux 2.069.1 runtime: user 0m0.053s <edited>
 Command: time dmd -deps=test.dep -c -o- test.d -de -w -m64 
 -color -g -debug -gs
 Linux runtime: user 0m0.584s
 OSX runtime: user 0m0.882s
Linux 2.069.1 runtime: user 0m0.035s <edited>
 Command: time dmd -c -o- test.d -de -w -m64 -color -g -debug -gs
 Linux runtime: user 0m0.048s
 OSX runtime: user 0m0.074s
Linux 2.069.1 runtime: user 0m0.036s <edited>
 Environment:
 OSX: El Capitan 10.11.6
 Linux with DMD 2.076.1: Gentoo 4.9.34
 Linux with DMD 2.069.1: Centos 7
Oct 10
prev sibling parent reply Jonathan M Davis <newsgroup.d jmdavisprog.com> writes:
On Wednesday, October 11, 2017 06:25:19 Dhananjay via Digitalmars-d-learn 
wrote:
 Hello,

 I am upgrading to DMD 2.076.1 from DMD 2.069.2 (similar results
 on 2.075.1), and seeing a huge increase in unittest compilation
 time when the -deps parameter is also passed to dmd. This is on
 both OSX and linux. What can be the cause of this?
Well, that's a pretty big version jump. So, a lot could have changed. One thing that comes to mind would be that imports were overhauled pretty thoroughly to try and fix various import bugs. This blog article talks about some of that: http://www.schveiguy.com/blog/2016/03/import-changes-in-d-2-071/ Or the change could be the result of something else entirely. Figuring it out would likely require doing a fair bit of debugging to narrow down when the change happened (and that's assuming that it's caused by a single commit or small set of commits rather than simply getting worse over time due to a variety of factors). - Jonathan M Davis
Oct 11
parent reply Saurabh Das <saurabh.das gmail.com> writes:
On Wednesday, 11 October 2017 at 08:11:37 UTC, Jonathan M Davis 
wrote:
 On Wednesday, October 11, 2017 06:25:19 Dhananjay via 
 Digitalmars-d-learn wrote:
 Hello,

 I am upgrading to DMD 2.076.1 from DMD 2.069.2 (similar 
 results on 2.075.1), and seeing a huge increase in unittest 
 compilation time when the -deps parameter is also passed to 
 dmd. This is on both OSX and linux. What can be the cause of 
 this?
Well, that's a pretty big version jump. So, a lot could have changed. One thing that comes to mind would be that imports were overhauled pretty thoroughly to try and fix various import bugs. This blog article talks about some of that: http://www.schveiguy.com/blog/2016/03/import-changes-in-d-2-071/ Or the change could be the result of something else entirely. Figuring it out would likely require doing a fair bit of debugging to narrow down when the change happened (and that's assuming that it's caused by a single commit or small set of commits rather than simply getting worse over time due to a variety of factors). - Jonathan M Davis
The following observations (for the above test program) were recorded on Mac OS X 10.11.6: DMD64 D Compiler v2.073.1 real 0m0.091s user 0m0.067s sys 0m0.020s DMD64 D Compiler v2.074.0 real 0m0.105s user 0m0.072s sys 0m0.022s DMD64 D Compiler v2.075.1 real 0m44.932s user 0m35.732s sys 0m7.098s DMD64 D Compiler v2.076.1 real 0m46.833s user 0m37.827s sys 0m7.254s Furthermore, 2.075.1 gave a bunch of deprecation warnings in std/string.d. The -de switch had to be removed before running. Hope this helps, Saurabh
Oct 13
parent reply Joakim <dlang joakim.fea.st> writes:
On Saturday, 14 October 2017 at 04:36:25 UTC, Saurabh Das wrote:
 On Wednesday, 11 October 2017 at 08:11:37 UTC, Jonathan M Davis 
 wrote:
 On Wednesday, October 11, 2017 06:25:19 Dhananjay via 
 Digitalmars-d-learn wrote:
 Hello,

 I am upgrading to DMD 2.076.1 from DMD 2.069.2 (similar 
 results on 2.075.1), and seeing a huge increase in unittest 
 compilation time when the -deps parameter is also passed to 
 dmd. This is on both OSX and linux. What can be the cause of 
 this?
Well, that's a pretty big version jump. So, a lot could have changed. One thing that comes to mind would be that imports were overhauled pretty thoroughly to try and fix various import bugs. This blog article talks about some of that: http://www.schveiguy.com/blog/2016/03/import-changes-in-d-2-071/ Or the change could be the result of something else entirely. Figuring it out would likely require doing a fair bit of debugging to narrow down when the change happened (and that's assuming that it's caused by a single commit or small set of commits rather than simply getting worse over time due to a variety of factors). - Jonathan M Davis
The following observations (for the above test program) were recorded on Mac OS X 10.11.6: DMD64 D Compiler v2.073.1 real 0m0.091s user 0m0.067s sys 0m0.020s DMD64 D Compiler v2.074.0 real 0m0.105s user 0m0.072s sys 0m0.022s DMD64 D Compiler v2.075.1 real 0m44.932s user 0m35.732s sys 0m7.098s DMD64 D Compiler v2.076.1 real 0m46.833s user 0m37.827s sys 0m7.254s Furthermore, 2.075.1 gave a bunch of deprecation warnings in std/string.d. The -de switch had to be removed before running. Hope this helps, Saurabh
I can reproduce on linux/x64, looks like a memory leak, as dmd balloons out to eat up all available memory until it's killed. I see it with this minimal command passed to dmd 2.075.1, but not the 2.074.1 frontend, as reported: ./dmd2/linux/bin64/dmd -c -o- foo.d -unittest -deps=foo.deps The closest issue I was able to find in bugzilla is this one, but that says it goes away with -o-, not the case here: https://issues.dlang.org/show_bug.cgi?id=17601 I suggest one of you file a bug with the minimal command, noting that it goes away if -unittest or -deps is not passed. Make sure you mark it as a regression, just like the above bug, as Walter pays special attention to those.
Oct 14
parent Saurabh Das <saurabh.das gmail.com> writes:
On Saturday, 14 October 2017 at 09:03:05 UTC, Joakim wrote:
 On Saturday, 14 October 2017 at 04:36:25 UTC, Saurabh Das wrote:
 [...]
I can reproduce on linux/x64, looks like a memory leak, as dmd balloons out to eat up all available memory until it's killed. I see it with this minimal command passed to dmd 2.075.1, but not the 2.074.1 frontend, as reported: ./dmd2/linux/bin64/dmd -c -o- foo.d -unittest -deps=foo.deps The closest issue I was able to find in bugzilla is this one, but that says it goes away with -o-, not the case here: https://issues.dlang.org/show_bug.cgi?id=17601 I suggest one of you file a bug with the minimal command, noting that it goes away if -unittest or -deps is not passed. Make sure you mark it as a regression, just like the above bug, as Walter pays special attention to those.
Filed a bug report: https://issues.dlang.org/show_bug.cgi?id=17898 Hope I didn't miss anything.
Oct 14