Charles-François:
> Also, I must admit I'm quite skeptical about the real benefit of
explicit probes for user-land, especially for CPython which isn't used
for performance-critical systems...

I beg to differ.  CPython is totally used on performance-critical systems, and I know I'm not the only user who thinks that.  "Performance-critical" doesn't necessarily mean "goes as fast as it ever possibly can", clearly PyPy is the place to go for that, but "can process at least X work in Y time".  Meeting performance goals with CPython is already challenging enough, please don't make it artificially hard by refusing to integrate tools which help users understand and improve performance.

Benjamin:
> I'm -1 on this patch for essentially the same reasons as Charles-François. It introduces a lot of code (and hacks!) in critical pathways of the interpreter. Someone would have to be constantly maintaining and testing it. In return, what do we get?

You get support for a highly sophisticated and low-impact profiling and tracing technology which provides support for illuminating performance problems *as well as* complicated behavioral problems that only happen under load, without slowing down the interpreter as a whole.  Not to mention possible integration with a whole slew of tools that know how to deal with data from that system.

I'm not saying that this is necessarily worth the maintenance burden; your analysis of the tradeoff may ultimately be correct.  I can't presume to know that because I am not intimately familiar with all the code it touches.  But it's definitely not nothing.