This is the mail archive of the binutils@sourceware.org mailing list for the binutils project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Undefined weak symbols in shared libraries


On Tue, 29 Jul 2008, Nick Clifton wrote:

> Hi Daniel,
> 
> > - The testsuite never runs C tests on cross targets.  This is an
> > annoying limitation for me since I mostly cross-test.  It's fixable,
> > but will introduce some dependencies on having a target during cross
> > testing that are not currently present.  Any opinions on that?
> 
> If you need a cross compiler why not put the test in the gcc testsuite ?
> 
> Answer: because it is testing the linker not the compiler.  But I think that
> the execution tests in the gcc testsuite should be considered to be toolchain
> tests rather then just compiler tests and hence it is reasonable to include
> checks on the linker's behaviour.  Besides there already are linker-feature
> tests in the gcc testsuite, so there is a precedent.

I'd suggest another possibility: a separate DejaGnu testsuite, in the src 
repository, for linker tests that use the linker only via the compiler and 
test that particular cases (with or without target-dependence) work 
correctly.

Such a testsuite would automatically work for multilibs and remote 
targets, by using whatever options are passed to runtest to configure 
compilation in the usual way.  The present ld and binutils testsuites are 
hard to use with multilibs; various tests assume things about being able 
to link objects built with the compiler and the assembler, or to link 
objects built with the compiler using particular linker options, and when 
the compiler options affect the assembler or linker options needed then 
the tests fail.  The effect is that you can typically test with default 
compiler options and whatever linker and assembler options the testsuite 
uses, but not with all the different multilib compiler options supported.

I would hope such tests could be shared with gold as far as possible, 
although tests checking the details of dumps would need to be conditioned 
in some way.

-- 
Joseph S. Myers
joseph@codesourcery.com


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]