I'm currently working through the dmd (and phobos for that matter), and come across it's regex implementation. Here I found out it makes rather heavy usage of malloc / pureMalloc; on one side to allocate the Matcher implementation inside GenericFactory; on the other side in BacktrackingMatcher for memory of it's stack.

Is there any historic reason why it dosnt uses more D-like features like new or dynamic arrays? It would certainly be safer to do so than using an untyped region of memory as stack-space.

Also: if it needs "plain" memory; would this code be a good candidate to utilize std.experimental.allocator once released?

Cheers,
Mai

On Thursday, 5 October 2023 at 13:36:10 UTC, Mai Lapyst wrote:

Not a formal answers but two clues:

1. when malloc is used in Phobos it's generally to bypass the GC
2. std.experimental.allocator was created way after std.regex

Maybe std.regex will see your question and answer more accurately. He was active a few weeks ago, coincidentally to annouce he had started working on a better design
for a D regex engine (https://forum.dlang.org/post/ctdkmvzttmpxwixkmvno@forum.dlang.org)