Re: [PATCH 2/2] gdb: use std::vector<> to hold on blocks in struct blockvector

[Simon Marchi <simark@simark.ca>]

On 2025-10-14 16:40, Jan Vraný wrote:
> On Tue, 2025-10-14 at 16:19 -0400, Simon Marchi wrote:
>> On 10/14/25 4:05 PM, Tom Tromey wrote:
>>>>>>>> "Simon" == Simon Marchi <simark@simark.ca> writes:
>>>
>>> Simon> I also wish we could avoid having a "set_block" method, because it
>>> Simon> requires having the blockvector in an invalid state (some blocks slots
>>> Simon> set to nullptr) while we build it, which could be error prone.  The
>>> Simon> users of the set_block method are buildsym and jit.  From what i saw,
>>> Simon> they could both very well prepare an std::vector with their blocks and
>>> Simon> std::move that vector into a blockvector constructor.  That constructor
>>> Simon> could assert that the blocks are correctly ordered, or to the sort
>>> Simon> itself (something that both buildsym and jit do already).  buildsym
>>> Simon> starts with a singly-linked list (m_pending_blocks), moves the blocks to a
>>> Simon> temporary vector (in end_compunit_symtab_get_static_block), sorts that
>>> Simon> vector, builds a singly-linked list again, and then traverses that list
>>> Simon> to insert the blocks into the blockvector.  I think that could all be
>>> Simon> simplified by making m_pending_blocks an std::vector<block *> from the
>>> Simon> start.
>>>
>>> Are you proposing that Jan do this in this patch?  Or is this just a
>>> future-looking comment?  It wasn't clear to me.
>>>
>>> Tom
>>
>> Sorry about that.  I would not ask Jan to refactor or clean up
>> everything in buildsym and jit.  But perhaps we can agree on what a good
>> (easy to use, hard to mis-use) API for blockvector is, and then Jan can
>> do the bare minimum to wire up buildsym and jit to that API.  We can
>> then do some cleanups later to remove some unnecessary steps in buildsym
>> and jit.
> 
> At the moment it seems to me that the first step is to drop add_block(),
> add append_block() and common ordering predicate and take it from there.
> I'll try to have a look tomorrow.

Sounds fine to me.  And I think it's fine to have set_block for now.
Code today pokes the block list directly, so set_block isn't introducing
anything new.

Simon