]> code.ossystems Code Review - openembedded-core.git/commit
icu: make filtered data generation optional, serial and off by default
authorAlexander Kanavin <alex.kanavin@gmail.com>
Tue, 23 Jun 2020 21:08:37 +0000 (23:08 +0200)
committerSteve Sakoman <steve@sakoman.com>
Sat, 11 Jul 2020 03:11:12 +0000 (17:11 -1000)
commitdf4bfb154e7fce9645f5738aa04b3eb3ef5093fe
tree4f5c880b4058464ed63493bda3e4cb289f41b0c9
parent80b61573fd1509cb46d263d6adb3621c3dee15d1
icu: make filtered data generation optional, serial and off by default

Backported from master with two differences:

1. Move exit in do_make_icudata_class-target after big endian patch
application (not required for master since new upstream icu includes
the patch)

2. In do_install_append_class-target test for existence of .dat input
file before calling icupkg

icu data generation was found to be racy, and causig AB failures;
making it serial and leaving it on is not an option as it regresses
to several minutes.

The specific bug is that rules.mk has:

LD_LIBRARY_PATH=../lib:../stubdata:../tools/ctestfw:$LD_LIBRARY_PATH  ../bin/gencnval -s . -d ./out/build/icudt66l mappings/convrtrs.txt

which creates a file and numerous rules like

LD_LIBRARY_PATH=../lib:../stubdata:../tools/ctestfw:$LD_LIBRARY_PATH  ../bin/genrb -s ./misc -d ./out/build/icudt67l -i ./out/build/icudt67l -k -q numberingSystems.txt

which quietly read it. There is no prerequisite for the former to complete first.

The race is extra complicated to fix as rules.mk is itself
generated through a custom in-tree python tool.

Signed-off-by: Alexander Kanavin <alex.kanavin@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
(cherry picked from commit df89e8d1136fd406ba35ae573e2cb0cfc88c6aad)
Signed-off-by: Steve Sakoman <steve@sakoman.com>
meta/recipes-support/icu/icu.inc
meta/recipes-support/icu/icu_66.1.bb