This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [PATCH 0/3] Introduce gdb::array_view, symtabs_and_lines -> std::vector
- From: Pedro Alves <palves at redhat dot com>
- To: GDB Patches <gdb-patches at sourceware dot org>
- Date: Mon, 4 Sep 2017 17:15:46 +0100
- Subject: Re: [PATCH 0/3] Introduce gdb::array_view, symtabs_and_lines -> std::vector
- Authentication-results: sourceware.org; auth=none
- Authentication-results: ext-mx02.extmail.prod.ext.phx2.redhat.com; dmarc=none (p=none dis=none) header.from=redhat.com
- Authentication-results: ext-mx02.extmail.prod.ext.phx2.redhat.com; spf=fail smtp.mailfrom=palves at redhat dot com
- Dmarc-filter: OpenDMARC Filter v1.3.2 mx1.redhat.com 2D25C6AAF8
- References: <1503343667-6790-1-git-send-email-palves@redhat.com>
On 08/21/2017 08:27 PM, Pedro Alves wrote:
> Looking at the "list and multiple locations" patch [1], I thought that
> it's a bit annoying/frustrating to have do remember to free the sals
> array that a symtabs_and_lines object wraps, with 'xfree (sals.sals)'
> or the equivalent make_cleanup(xfree, ...), not to mention how
> error-prone manual memory management is. We can do better in C++.
>
> So I wrote a patch to replace 'struct symtabs_and_lines' with 'typedef
> std::vector<symtab_and_line> symtabs_and_lines'. That works nicelly,
> except that I wasn't entirely happy that that would introduce heap
> allocations in a few cases where there is none nowadays.
>
> I then addressed that by adding an array_view abstraction, which is
> something that I've wanted at other times before.
>
> After that, we'd end up with some places using a std::vector disguised
> behind the symtabs_and_lines typedef, while in other places we'd be
> spelling out gdb::array_view. With that, I no longer see any value in
> keeping the symtabs_and_lines typedef, kind of giving preference to
> std::vector, so I removed the typedef, choosing to spell out
> std::vector explicitly.
>
> (I'm aware that there are several proposals for adding something like
> array_view to the standard C++ library. In this implementation, which
> was written from scratch and developed in parallel with the unit
> tests, I chose to keep it as simple and safe as possible. For
> example, there is no support for ranks != 1, simply because the main
> use case this aims at addressing is abstracting out std::vector vs
> stack arrays in APIs.)
>
> [1] https://sourceware.org/ml/gdb-patches/2017-07/msg00280.html
>
> Note that the net increase shown below is mostly caused by the new
> unit tests. If we only count patches 2 and 3, then we get instead:
>
> 33 files changed, 537 insertions(+), 820 deletions(-)
FYI, I'm pushing this in now.
Thanks,
Pedro Alves