Broken inlining

[yuyichao]

Seems that inlining is broken on the current master.

Tested using the test script for #13286 .

For bad commit, the code_typed clearly shows that many simple functions are not inlined.

Bisected to a33cd07

a33cd07f8a0fe2dd4ef3fecf809add2bb2a493c9 is the first bad commit
commit a33cd07f8a0fe2dd4ef3fecf809add2bb2a493c9
Author: Jameson Nash <vtjnash@gmail.com>
Date:   Wed Oct 7 16:35:53 2015 -0400

    preserve inline line numbers during inlining

:040000 040000 d51f01700bc9077c02bb204a541e6746967a4617 423dc659aee3291be27807b8f0f23cab918241d1 M        base

Bisect log

git bisect start
# bad: [af71b86edde26e607fc95fd8a8e73075ef675745] Do not pass RTLD_DEEPIND to dlopen in asan mode
git bisect bad af71b86edde26e607fc95fd8a8e73075ef675745
# good: [b5bebf2b05f63ad11f19a252cff2244a719337e8] Merge pull request #13453 from matthieugomez/patch-1
git bisect good b5bebf2b05f63ad11f19a252cff2244a719337e8
# bad: [646ea37c810c0f6e8c099b3d7f1c45f96b66fe6d] Merge pull request #13491 from JuliaLang/jn/lineno_fixes
git bisect bad 646ea37c810c0f6e8c099b3d7f1c45f96b66fe6d
# good: [930e6b55327c6caf60d0eae4d71ffdf43e55fbae] set a min target architecture of pentium4 to require support for sse2 and _mm_pause
git bisect good 930e6b55327c6caf60d0eae4d71ffdf43e55fbae
# good: [fb4b161ece65a6e48451f97aa2bfd0ba986b2709] Merge pull request #13478 from wildart/fix-13375
git bisect good fb4b161ece65a6e48451f97aa2bfd0ba986b2709
# good: [0d923f917d2269f25b9173ca56c9fce06542ed33] Merge pull request #13489 from KristofferC/kc/amulbdoc
git bisect good 0d923f917d2269f25b9173ca56c9fce06542ed33
# bad: [af81d6c7fe09f57ab268ccd731bc2d2f4aee9a95] fix #922 (wrong line numbers in error expressions) and remove old hacks around this issue
git bisect bad af81d6c7fe09f57ab268ccd731bc2d2f4aee9a95
# bad: [a33cd07f8a0fe2dd4ef3fecf809add2bb2a493c9] preserve inline line numbers during inlining
git bisect bad a33cd07f8a0fe2dd4ef3fecf809add2bb2a493c9
# first bad commit: [a33cd07f8a0fe2dd4ef3fecf809add2bb2a493c9] preserve inline line numbers during inlining

Noticed that this is already caught by @andreasnoack in the commit comment after the bisect. I decided to post it anyway since there doesn't seems to be an issue for it yet.

cc @vtjnash

[damiendr]

In case you need an easier way to detect this with automated tools: the code I posted (#13552) incurs a very severe penalty, 9x slower and 30x more memory allocations.