[Bug mi/29426] FAIL: gdb.mi/mi-var-invalidate.exp: linteger not anymore in scope due to binary changes (unexpected output)

["vries at gcc dot gnu.org" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=29426

--- Comment #7 from Tom de Vries <vries at gcc dot gnu.org> ---
(In reply to Lancelot SIX from comment #6)
> Hi
> 
> This is what I also see.  TBH, I am really wondering why I have not seen
> with my testing and with my ubuntu boxes.
> 

It's PIE vs. no-PIE (and not for the first time ;) ).

So my system has no-PIE by default (because I haven't installed the gcc-PIE
package).

Using target board unix/-fPIE/-pie (case PIE), the test passes.  Using unix
(case no-PIE), it fails.

On ubuntu, there's a PIE default for system gcc.  When you build gcc from
scratch, it probably had no-PIE default instead.

Anyway, in one case (no-PIE) var_create succeeds, in the other (PIE) not.

So, varobj_create calls get_frame_block, which calls block_for_pc, with a
relocated pc.  Then block_for_pc calls block_for_pc_sect, which call
blockvector_for_pc_sect, which calls find_pc_sect_compunit_symtab, which calls
objf->find_pc_sect_compunit_symtab with objf == mi-var-invalidate_bis.

In the PIE case, this query returns null because we try to find an relocated
address in an unrelocated objfile.

In the no-PIE case, this query returns non-null because we try to find an
non-relocatable address in an non-relocatable objfile.


In the PIE case

> I am working on a fix for this.
> 
> Thanks for taking the time and looking into it.

-- 
You are receiving this mail because:
You are on the CC list for the bug.