Re: [PATCH 1/2] gdb: allocate blockvector on heap

[Simon Marchi <simark@simark.ca>]

On 10/13/25 2:23 PM, Jan Vrany wrote:
> This patch changes blockvector to be allocated on the heap, using
> 'new'.
> 
> The code is little awkward since blockvector uses flexible array member
> to hold on blocks. So in order to make it working, this commit overloads
> 'new' and 'delete' operators to take the number of blocks as parameter
> and allocates/deallocates memory using std::malloc and std::free.
> This will go away in following commit that will change blockvector to
> use std::vector rather than flexible array member.

It's already written, so if it works, I don't mind, but otherwise I
think you could have made the switch to 'new' and to std::vector in the
same patch, since they are not very large changes.

> ---
>  gdb/block.h      | 18 ++++++++++++++++++
>  gdb/buildsym.c   |  5 +----
>  gdb/jit.c        |  8 +-------
>  gdb/mdebugread.c | 38 +++++++++++---------------------------
>  4 files changed, 31 insertions(+), 38 deletions(-)
> 
> diff --git a/gdb/block.h b/gdb/block.h
> index 8a3ea822d04..d1ca41d7482 100644
> --- a/gdb/block.h
> +++ b/gdb/block.h
> @@ -420,6 +420,24 @@ struct global_block : public block
>  
>  struct blockvector
>  {
> +  void *operator new (size_t size, size_t num_blocks)
> +  {
> +    size_t size_in_bytes = size + (num_blocks - 1) * (sizeof (struct block *));
> +    return std::malloc (size_in_bytes);
> +  }
> +
> +  void operator delete (void *memory)
> +  {
> +    std::free (memory);
> +  }
> +
> +  blockvector (int nblocks)

explicit

> +    : m_map (nullptr),

Can you initialize m_map inline?

  addrmap_fixed *m_map = nullptr;

Otherwise, LGTM.

Approved-By: Simon Marchi <simon.marchi@efficios.com>

Simon