Re: [PATCH] pkl: Get rid of string literals as exception names

[Mohammad-Reza Nabipoor]

On Sun, Jan 30, 2022 at 10:07:16PM +0100, Luca Saiu wrote:
> On 2022-01-30 at 15:55 +0330, Mohammad-Reza Nabipoor wrote:
> >
> > Yet another reason for not spoting this problem until recently, is my 
> > workflow;
> > in which I only use switch-dispatching. Because it's very 
> > debug/debugger-friendly.
> > (I'm not sure about Jose).
> 
> You are right about this, and I will document this as a suggestion for
> development.  Direct-threading should be just as friendly, even if it
> depends on GNU C -- minimal-threading and no-threading are definitely
> not friendly to gdb.
> 

What about supporting all supported dispatches in a single library?
Currently there are `libjitter--{switch,{direct,minimal,no}-threading}`
libraries. All providing the same functions. If you add a suffix to all
symbols, they can live inside a single `libjitter` and for 
backward-compatibility
you can provide current symbols as a synonym for the best dispatch available.
Either by macros, or linker tricks, or real functions that call corresponding
best dispatch functions.

I don't know how implementable this approach is. But from my non-expert
point of view, it seems doable.

Then applications, like poke, can provide, e.g. on x86_64, four executables,
poke (no-threading), poke_minimal-threading, poke_direct-threading,
poke_switch-threading.

This way, I think, testing the different dispatches become simpler.
It can makes life for CI easier to run the tests for more architectures.
Thanks to people, like Romain, we have access to more tests of poke (and jitter)
on more platforms, easier.

WDYT?

Jose, WDYT?


Regards,
Mohammad-Reza