Guard compile tests from running when unsupported + harden feature support check

  In our test environment i noticed a few compile tests, though unsupported,
were still being executed and causing spurious FAIL's.

The following change adds a missing feature support check to compile-ifunc.exp
and also adds one more pattern to the feature support check function.

With this change, none of the compile tests run when they're not supposed to.

OK?

gdb/testsuite/ChangeLog:

2015-08-19  Luis Machado  <lgustavo@codesourcery.com>

	* lib/gdb.exp (skip_compile_feature_tests): Check for unsupported
	compiler language.
	* gdb.compile/compile-ifunc.exp (with_test_prefix): Check feature
	support before running tests.
---
 gdb/testsuite/gdb.compile/compile-ifunc.exp | 5 +++++
 gdb/testsuite/lib/gdb.exp                   | 3 +++
 2 files changed, 8 insertions(+)
  

On 08/19/2015 01:53 PM, Luis Machado wrote:

> +    if {[skip_compile_feature_tests]} {
> +	untested "compile command not supported (could not find libcc1 shared library?)"
> +	return -1
> +    }
> +
>      # gnu_ifunc (10): error: too many arguments to function 'gnu_ifunc'
>      gdb_test_no_output "compile code resultvar = gnu_ifunc_alias (10);"
>  

This one's OK.

> diff --git a/gdb/testsuite/lib/gdb.exp b/gdb/testsuite/lib/gdb.exp
> index 56cde7a..fcf9cac 100644
> --- a/gdb/testsuite/lib/gdb.exp
> +++ b/gdb/testsuite/lib/gdb.exp
> @@ -3005,6 +3005,9 @@ proc skip_compile_feature_tests {} {
>  	-re "Command not supported on this host\\..*\r\n$gdb_prompt $" {
>  	    set result 1
>  	}
> +	-re "No compiler support for this language\\.\r\n$gdb_prompt $" {
> +	    set result 1
> +	}

This one I'm not so sure.

I'd suspect that that could happen when e.g., gdb connects to gdbserver and
finds the program stopped at the entry point, and then the current language
ends up set to asm instead of C.  But then that would be a test bug and it
would be wrong to skip further testing.

But please don't make us guess; please expand on when do you see this.

Thanks,
Pedro Alves
  

On 08/19/2015 10:15 AM, Pedro Alves wrote:
> On 08/19/2015 01:53 PM, Luis Machado wrote:
>
>> +    if {[skip_compile_feature_tests]} {
>> +	untested "compile command not supported (could not find libcc1 shared library?)"
>> +	return -1
>> +    }
>> +
>>       # gnu_ifunc (10): error: too many arguments to function 'gnu_ifunc'
>>       gdb_test_no_output "compile code resultvar = gnu_ifunc_alias (10);"
>>
>
> This one's OK.
>
>> diff --git a/gdb/testsuite/lib/gdb.exp b/gdb/testsuite/lib/gdb.exp
>> index 56cde7a..fcf9cac 100644
>> --- a/gdb/testsuite/lib/gdb.exp
>> +++ b/gdb/testsuite/lib/gdb.exp
>> @@ -3005,6 +3005,9 @@ proc skip_compile_feature_tests {} {
>>   	-re "Command not supported on this host\\..*\r\n$gdb_prompt $" {
>>   	    set result 1
>>   	}
>> +	-re "No compiler support for this language\\.\r\n$gdb_prompt $" {
>> +	    set result 1
>> +	}
>
> This one I'm not so sure.
>
> I'd suspect that that could happen when e.g., gdb connects to gdbserver and
> finds the program stopped at the entry point, and then the current language
> ends up set to asm instead of C.  But then that would be a test bug and it
> would be wrong to skip further testing.
>

No stopping at the entry point, we run to main.

> But please don't make us guess; please expand on when do you see this.

It is related to how the binaries are built, without debug info. That's 
how GDB reacts to these GCC 5.2-produced binaries.

compile-ifunc.exp:

#0  0x100007b0 in main ()
(gdb) show language
show language
The current source language is "auto; currently asm".

compile-ops.exp:

Breakpoint 1, 0x10000724 in func (param=268503652, 
optimized_out=<optimized out>)^M
(gdb) compile code -- ;^M
No compiler support for this language.^M

show language
The current source language is "auto; currently asm".

Enabling debug info i see the standard error about missing libcc1, which 
may be an alternate fix, though checking for unsupported languages does 
sound useful to me.
  

On 08/19/2015 02:34 PM, Luis Machado wrote:

>> But please don't make us guess; please expand on when do you see this.
> 
> It is related to how the binaries are built, without debug info. That's 
> how GDB reacts to these GCC 5.2-produced binaries.
> 

...

> 
> Enabling debug info i see the standard error about missing libcc1, which 
> may be an alternate fix, though checking for unsupported languages does 
> sound useful to me.

I disagree.  It's a bug, because the tests are testing C code.

> compile-ifunc.exp:
>
> #0  0x100007b0 in main ()
> (gdb) show language
> show language
> The current source language is "auto; currently asm".
>

This binary has no debug info (by test design), so we should just
make the test do "set language c" (with a comment).

> compile-ops.exp:
>
> Breakpoint 1, 0x10000724 in func (param=268503652,
> optimized_out=<optimized out>)^M
> (gdb) compile code -- ;^M
> No compiler support for this language.^M
>
> show language
> The current source language is "auto; currently asm".

This one looks like a gdb bug?  AFAICS, the test generates
debug info, and gdb should have figured out the language is c.
At least, from skimming the test's .exp, I see:

	compile_unit {
	    {name file1.txt}
	    {language @DW_LANG_C}
             ^^^^^^^^^^^^^^^^^^^
	    {low_pc func_start addr}
	    {high_pc func_end addr}
	} {
	    global program


BTW, it'd be nice if the "No compiler support for this language"
error told us which language is "this"...

Thanks,
Pedro Alves
  

On 08/19/2015 10:45 AM, Pedro Alves wrote:
> On 08/19/2015 02:34 PM, Luis Machado wrote:
>> Enabling debug info i see the standard error about missing libcc1, which
>> may be an alternate fix, though checking for unsupported languages does
>> sound useful to me.
>
> I disagree.  It's a bug, because the tests are testing C code.
>

I did a bit more investigation and here's what i found:

GDB generally sets the default language to C on top.c:gdb_init. Usually 
the dynamic loader code is not available, so GDB doesn't switch away 
from C when connecting to GDBserver.

It does for me because it sees real asm code from the dynamic loader, so 
the starting language is "asm". Now, the testcases have different behaviors.

For compile-ifunc.exp, when we run to main, we don't have any kind of 
debug info that tells GDB what language we are dealing with, so GDB 
doesn't switch away from the previously selected "asm". Forcing the 
language to C would fix this.

For compile-ops.exp, on the other hand, we run to "func" and it has 
debug info, but the source file is named "file1.txt". Not sure if this 
was on purpose, but it seems to throw GDB off when trying to find the 
source language, even though the DIE states it is C.

If i rename the source file to file1.c, then GDB correctly selects C as 
the current source language.

It could be either a GDB bug for not honoring the language in the DIE 
itself or a testcase issue for not naming the source file with the 
correct language extension.

> BTW, it'd be nice if the "No compiler support for this language"
> error told us which language is "this"...
>

Indeed it would be useful.