[PATCH V5] CTF: multi-CU and archive support

[Lancelot SIX]

> Hi,
> 
> You can find the latest gcc patch series for CTF support, which is currently
> under review, here:
> 
> http://gcc.1065356.n8.nabble.com/PATCH-V9-0-7-Support-for-the-CTF-and-BTF-debug-formats-td1897109.html
> 
> Also please make sure to use a binutils version with CTF support.
> 

Hi,

Thanks for the pointer.

Now that I have a ctf-aware gcc, I can give a first try to the patch.

It looks like the flag to enable CTF generation (at least in this
version of the GCC patch set) is '-gctf', not '-gt'.

The '-gt' flag is currently used in the tests to detect if the
compiler supports CTF (in gdb/testsuite/lib/gdb.exp).  In the current
situation, tests are skipped with GCC master + CTF.  We might want to
wait for GCC to settle on the flag, and update tests codebase as well as
this patch.

> 
> I believe  the psymtab_discarder's dtor should take care of discarding all
> built partial
> symbol tables. So I'm not sure if this is a CTF specific issue.
> 

It is taken care of by psymtab_storage::~psymtab_storage which is
eventually triggered by the dtor of objfile.  I did not dig enough
before sending the email.  Sorry for that.

> One way to avoid dealing with either the storage or dtor issue is to
> declare:
> 
>         struct ctf_context context;
> 
> in struct ctf_psymtab.

Yes, this would work.  However, considering what I missed with the
memory management of the psymtab and given that ctf_context is a POD
type, everything should properly come down with the dtor of objfile.  So
after reconsidering I believe your proposition should works.

Lancelot.