Re: [PATCH v3 1/2] gdb/testsuite: Add libc_has_debug_info require helper

[Pedro Alves <pedro@palves.net>]

On 2024-04-26 04:00, Thiago Jung Bauermann wrote:
> 
> Hello Bernd,
> 
> Bernd Edlinger <bernd.edlinger@hotmail.de> writes:
> 
>> Hi Thiago,
>>
>> On 4/24/24 18:25, Thiago Jung Bauermann wrote:
>>>>
>>> Thank you! Pushed as commit f5ef12c3f1af.
>>
>> I think I have an issue with this commmit.
>> I use a self-built riscv-unknown-elf toolchain with newlib,
>> so there is no libc at all, regardless of debug info.
>> since today, I see messages like:
>> Running /home/ed/gnu/binutils-build-riscv64/gdb/testsuite/../../../binutils-gdb/gdb/testsuite/gdb.base/relativedebug.exp ...
>> FAIL: gdb.base/relativedebug.exp: info sharedlibrary libc.so
>> ERROR: tcl error sourcing /home/ed/gnu/binutils-build-riscv64/gdb/testsuite/../../../binutils-gdb/gdb/testsuite/gdb.base/relativedebug.exp.
>> ERROR: tcl error code TCL READ VARNAME
>> ERROR: can't read "libc_has_debug_info": no such variable
>>     while executing
>> "verbose "$me: returning $libc_has_debug_info" 2"
>>     (procedure "gdb_real__libc_has_debug_info" line 47)
>>     invoked from within
>> "gdb_real__libc_has_debug_info"
> 
> <snip>
> 
> Sorry for the trouble. I should have simulated a situation where GDB
> can't find libc.so in the inferior. I was able to reproduce the error
> above when I did.
> 
> Could you please test the patch that I just sent?
> 
>> while previously that looked like:
>>
>> gdb compile failed, /home/ed/gnu/riscv64-unknown-elf/lib/gcc/riscv64-unknown-elf/14.0.1/../../../../riscv64-unknown-elf/bin/ld: /tmp/ccjr19GC.o: in function `main':
>> /home/ed/gnu/binutils-build-riscv64/gdb/testsuite/../../../binutils-gdb/gdb/testsuite/gdb.base/relativedebug.c:30:(.text+0x28): undefined reference to `alarm'
>> /home/ed/gnu/riscv64-unknown-elf/lib/gcc/riscv64-unknown-elf/14.0.1/../../../../riscv64-unknown-elf/bin/ld:
>> /home/ed/gnu/binutils-build-riscv64/gdb/testsuite/../../../binutils-gdb/gdb/testsuite/gdb.base/relativedebug.c:31:(.text+0x30):
>> undefined reference to `pause'
>> /home/ed/gnu/riscv64-unknown-elf/lib/gcc/riscv64-unknown-elf/14.0.1/../../../../riscv64-unknown-elf/bin/ld:
>> /home/ed/gnu/binutils-build-riscv64/gdb/testsuite/../../../binutils-gdb/gdb/testsuite/gdb.base/relativedebug.c:31:(.text+0x38):
>> undefined reference to `pause'
>> collect2: error: ld returned 1 exit status
>> UNTESTED: gdb.base/relativedebug.exp: failed to compile
>>
>> so not very noisy, newlib does apparently not have alarm, pause, sleep, and similar,
>> but much easier to understand the output...
> 
> On the plus side, with this problem fixed gdb.base/relativedebug.exp
> should exit early with:
> 
> (gdb) info sharedlibrary libc.so
> No shared libraries matched.
> (gdb) UNSUPPORTED: gdb.base/relativedebug.exp: require failed: libc_has_debug_info (libc not found in the inferior)
> 
> Which will be even easier to understand the output. :-)
> 

I don't think that's a good outcome, actually.  It'll disable the testcase on systems that link
with their libc statically (even if has debug info), or systems that name their libc something else.
And I worry that the require predicate will start being used more with that particularity.

The original check only returned early if a library called "*libc*" was found, and, it didn't
have debug info.  If those conditions didn't match, the testcase proceeded.  That seems a
lot safer.  As in, only if we know for sure we have a libc without debug info, do we return early.