Bug #16772: Build becomes slow with CIFS mounted srcdir

• Status: Closed
• Assignee: shyouhei (Shyouhei Urabe)
• Target version: -
• Backport: 2.5: UNKNOWN, 2.6: UNKNOWN, 2.7: UNKNOWN

[ko1 (Koichi Sasada)]

Recent 9e6e39c3512f7a962c44dc3729c98a0f8be90341 increases many include files and it makes slow build on my machine.

My build environment is not normal:

• srcdir is managed on Windows 10 host file system and edit it on windows native editors (xyzzy)
• Build on Linux (Ubuntu 18.04) by mounting srcdir with CIFS (samba)

# on e7128aa772787a9c9e95708e2d2534484861a765
# gcc version 7.5.0 (Ubuntu 7.5.0-3ubuntu1~18.04)
# on Ubuntu 18.04
# on Hyper-V
# on Windows 10
# srcdir is mounted by CIFS (Samba) on Windows host file system

# without ccache
$ time make range.o
compiling /home/ko1/src/ruby/clean/range.c

real    0m17.662s
user    0m2.572s
sys     0m2.599s

# with ccache
$ time make range.o
compiling /home/ko1/src/ruby/clean/range.c

real    0m35.361s
user    0m4.628s
sys     0m5.283s

on 9e6e39c3512f7a962c44dc3729c98a0f8be90341~

# without ccache
$ rm range.o; time -p make range.o
compiling /home/ko1/src/ruby/clean/range.c
real 1.16
user 0.85
sys 0.06

Same machine, but srcdir on guest Linux (/dev/sda2 on / type ext4 (rw,relatime,data=ordered))

# on e7128aa772787a9c9e95708e2d2534484861a765
$ time make range.o
compiling /home/ko1/ruby/v2/src/trunk/range.c

real    0m1.207s
user    0m1.065s
sys     0m0.140s

Same machine, but compile on host Windows (mswin64).

$ time nmake range.obj

Microsoft(R) Program Maintenance Utility Version 14.14.26428.1
Copyright (C) Microsoft Corporation.  All rights reserved.

generating id.h
id.h updated
compiling C:/ko1/src/ruby/clean/range.c
range.c

real    0m3.779s
user    0m0.000s
sys     0m0.046s

Another Linux box works fine.

# Linux box on another machine without ccache

$ time make range.o
compiling /home/ko1/ruby/v2/src/trunk/range.c

real    0m1.992s
user    0m1.858s
sys     0m0.123s

I recognize that using CIFS for build environment is not normal environment, so it is not a big issue.
But I filed it if there is a chance to solve it.

BTW, I tried to make simple .c files which includes 1000 simple files, but only a few seconds to build. So the file number is not a big issue.

[usa (Usaku NAKAMURA)]

On my Windows PC, building ruby without tests takes over 20 min now.
And mswin-CI (http://mswinci.japaneast.cloudapp.azure.com/vc12-x64/ruby-master/recent.html) also fails by timeout.

[shyouhei (Shyouhei Urabe)]

• Assignee set to shyouhei (Shyouhei Urabe)

[shyouhei (Shyouhei Urabe)]

• Related to Bug #16770: Commit: RUBY3_HAS_BUILTIN: fix for nonexistent builtin - large increase in MinGW & mswin compile times added

[Eregon (Benoit Daloze)]

@shyouhei (Shyouhei Urabe) maybe it was split in too many header files?
Agreed before ruby/ruby.h and intern.h were too big, but 186 headers vs 26 before seems the other extreme.

[ko1 (Koichi Sasada)]

• Status changed from Open to Closed

Applied in changeset git|443389effc37308ce1a3c3a840082a344fc6af56.

---

reduce duplicate include.

Without this patch, 20k files are opened (openat syscall) because
of duplicate includes. This patch reduced it to 3k and build time
was reduced compile time of range.o from 15sec -> 3sec on my machine.
[Bug #16772]