Re: [PATCH v2] [gdb/symtab] Fix dwarf version of DWO TU

[Simon Marchi <simark@simark.ca>]

On 1/21/25 10:09 AM, Tom de Vries wrote:
> On 1/21/25 14:32, Tom Tromey wrote:
>>>>>>> "Tom" == Tom de Vries <tdevries@suse.de> writes:
>>
>> Tom> When running test-case gdb.ada/access_tagged_param.exp with target board
>> Tom> fission, we run into:
>> Tom> ...
>> Tom> (gdb) break pck.adb:19^M
>> Tom> gdb/dwarf2/read.h:289: internal-error: version: \
>> Tom>   Assertion `m_dwarf_version != 0' failed.^M
>> Tom> ...
>>
>> Tom> -      && (cu->per_cu->version () == 2
>> Tom> +      && (cu->header.version == 2
>>
>> All the stuff in this area seems kind of horrible to me.  Like, first of
>> all, how can this even fail?  It seems like the CU version should be set
>> when setting up the reader. 
> 
> Hi Tom,
> 
> thanks for the review.
> 
> The v1 ( https://sourceware.org/pipermail/gdb-patches/2024-October/212676.html ) takes the approach of making sure "cu->per_cu->version ()" is set.
> 
> Do you prefer that one?
> 
> Thanks,
> - Tom
> 
>> And if not, what's gone wrong there?
>> Relatedly, can the other callers of version() fail in the same way?
>> Finally there aren't really that many callers of this method so I wonder
>> if it can be removed, maybe making this code less fragile.  Looking at
>> it, I feel pretty sure if I needed a version check I'd probably write
>> code to call this method but ... maybe that's unsafe?
>>
>> Tom
> 

My understanding of what Tom T said is that the
dwarf2_per_cu_data::version method could be removed.  Callers could use
the dwarf2_cu::header::version field, since they do have access to a
dwarf2_cu anyway.  I will give that a shot.

Simon