RE: Interests in a breakpoint hook/breakpoint op patch?

[Dan Tull <dtull@...>]

Thomas Jericke
> In the linked patch a function has to be on the stack to be able to 
> install a breakpoint. I don't have this requirement. As I need to be 
> able to add a breakpoint to a "running" interpreter at any time the 
> approach in the linked patch will not work for me. Instead I have to 
> have a structure of all protos running in the interpreter (currently I 
> use a linked list) to find the proto of a given source name and line number.

FWIW, the full version does support applying breakpoints to a proto even 
if if no live functions* exist that use it. That's the additional change to which 
this remark refers: 

> == Disadvantages ==
> - You have to be able to locate a reference to the function in which
>   you want to set the breakpoint. We actually made another mod
>   to make this faster and easier but that's a separate change.
(see: http://thread.gmane.org/gmane.comp.lang.lua.general/68867/focus=69271)

What we did sounds similar to your approach. It was added very shortly 
after I generated the patch, but I never went back and updated it (mostly
because nobody came asking).

There's an extra API lua_setprotohook that allows us provide a callback.
We use that to keep an up to date structure with all the protos so they can 
be patched or unpatched quickly without a function reference.

We've gotten a lot of benefit from having fast breakpoints. I've actually got
some ideas for using it for low overhead dtrace-like performance analysis and
fast code coverage, but haven't pursued those in earnest yet.

DT

* for example, if there's function exists returns closures, but it hasn't been
called at the time a breakpoint is set in it (or has been called, but all the results
were already garbage collected)