'make clean' kills libmbfl build about hhvm-third-party

[jwatzman]

@JoelMarcey and I were just looking at another issue with eaw_table.h -- the cmake build does regenerate it, except it wouldn't do it properly if you didn't have gawk installed. Joel is fixing that (require gawk), we should probably remove this file altogether?

[fredemmott]

fred@hhvm-oss-bench:~$ gawk --version | head -n 1
GNU Awk 4.0.1
fred@hhvm-oss-bench:~$ awk --version | head -n 1
GNU Awk 4.0.1

(on Ubuntu 14.04)

[JoelMarcey]

Does the fix not work then, @fredemmott ?

[fredemmott]

Trying now - however, it sounded like it was expected to work with gawk previously - it didn't.

[fredemmott]

Still broken - make -C third-party/libmbfl clean; make

re-running 'cmake' fixes this - but it should be built by 'make' if 'make clean' removes it.

[JoelMarcey]

@fredemmott I don't have an Ubuntu machine anymore. Is there any output you can share?

[fredemmott]

fred@hhvm-oss-bench:~/hhvm$ make -C third-party/libmbfl clean
make: Entering directory `/home/fred/hhvm/third-party/libmbfl'
make[1]: Entering directory `/home/fred/hhvm'
make[2]: Entering directory `/home/fred/hhvm'
make[2]: Leaving directory `/home/fred/hhvm'
make[1]: Leaving directory `/home/fred/hhvm'
make: Leaving directory `/home/fred/hhvm/third-party/libmbfl'
fred@hhvm-oss-bench:~/hhvm$ make
[  2%] Built target pcre
[  2%] Built target pcre_jit_test
[  2%] Built target pcrecpp
[  2%] Built target pcre_scanner_unittest
[  2%] Built target pcre_stringpiece_unittest
[  2%] Built target pcrecpp_unittest
[  2%] Built target pcreposix
[  2%] Built target pcregrep
[  3%] Built target pcretest
[  3%] Built target fastlz
[  3%] Built target afdt
[  3%] Building C object third-party/libmbfl/mbfl/CMakeFiles/mbfl.dir/mbfilter.c.o
/home/fred/hhvm/third-party/libmbfl/mbfl/mbfilter.c:107:23: fatal error: eaw_table.h: No such file or directory
 #include "eaw_table.h"
                       ^
compilation terminated.
make[2]: *** [third-party/libmbfl/mbfl/CMakeFiles/mbfl.dir/mbfilter.c.o] Error 1
make[1]: *** [third-party/libmbfl/mbfl/CMakeFiles/mbfl.dir/all] Error 2
make: *** [all] Error 2

[jwatzman]

Oh, I see, the code to generate the file is run as part of the cmake configuration, but it's not part of an actual build rule. I think we can just remove this line which adds it to make clean, right? https://github.com/hhvm/hhvm-third-party/blob/master/libmbfl/mbfl/CMakeLists.txt#L7 Or should we go thru the work of making it an actual build target?

[fredemmott]

lines 9-13 feel like they should be a build target that the rest depends on in an ideal world.

[kevinxucs]

(On Ubuntu 14.04)

After I run cmake (on a clean git clone), it generates a empty eaw_table.h, and the build fails.

When I try to manually run the generation command

awk -v TABLE_NAME=mbfl_eaw_table -f mk_eaw_tbl.awk EastAsianWidth.txt > eaw_table.h

It has following errors:

$ awk -v TABLE_NAME=mbfl_eaw_table -f mk_eaw_tbl.awk EastAsianWidth.txt > eaw_table.h
awk: mk_eaw_tbl.awk: line 83: function strtonum never defined
awk: mk_eaw_tbl.awk: line 83: function strtonum never defined
awk: mk_eaw_tbl.awk: line 83: function strtonum never defined

[JoelMarcey]

@kevinxucs We need to look into this. But when manually running the generation command, can you try using gawk instead of awk? You may get the same error, but I am curious.

[kevinxucs]

@JoelMarcey Ah... I found the problem, awk was linked to mawk, after I installed and switched to gawk it works fine.

[JoelMarcey]

@fredemmott @jwatzman The older mbfl/CMakeLists.txt before the libmbfl update did not have the clean line in it. And it also only generated the file if EastAsianWidth.txt wasn't available. But we had included that file in the repo along with eaw_table.h.

https://github.com/hhvm/hhvm-third-party/blob/1bb8025d4fc52188454c18b9df6a764dedd8bcb9/libmbfl/mbfl/CMakeLists.txt

Bottom line I think we can just remove the clean line and we might be good.

Agree?

[jwatzman]

It depends on how you see this eaw_table.h as part of the build. Is it an artifact of configuration (i.e., cmake), like the actual makefiles, in which case we should just remove the clean line. Or is it a build artifact, like object files, in which case we should make it a real build target?

I don't feel strongly either way. It probably doesn't really matter. Removing the clean line is the easiest I think.

(Its historical status isn't terribly relevant, since the existing makefile is pretty clearly broken anyways.)