[PATCH] Fix libLTO.so -disable-opt and other gold plugin options recently moved to lto.cpp

[Tom Roeder]

I just discovered that my recent patch (the one that ended up as
r191680) that moved some options around in the LTO code permanently
disabled these options for the gold plugin. The options affected are
the ones that were moved to tools/lto/lto.cpp: -disable-opt,
-disable-inlining, and -disable-gvn-loadpre.

The problem was that the command-line option parsing wasn't happening
until just before code generation in lib/LTO/LTOCodeGenerator.cpp, and
this was too late for the options to be set, since they are now passed
down from lto.cpp. So, this patch pulls out the command-line parsing
into a separate function in LTOCodeGenerator and calls it right before
code generation or merged module generation.

However, I don't have a test case for this; llvm-lto wasn't broken by
the change; its options work correctly. And we don't currently have a
test directory for LLVMgold.so or any regression tests that I know of.
This is probably because LLVMgold.so depends on the gold plugin, hence
on binutils, which isn't guaranteed to be present. But this bug
suggests that there needs to be some way to track regressions for
LLVMgold.so.

I suppose the right way to do this is to write tests conditional on
the presence of gold and put them in test/tools/gold. Are there
options in the test framework that could be used to select for this?

Thanks,

Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: lto_parse_options.patch
Type: application/octet-stream
Size: 3386 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20131001/01fb7e01/attachment.obj>