On Tue, 24 Mar 2026, Jeffrey Law wrote:

> 
> 
> On 3/24/2026 3:02 AM, Dongsheng Song via Gcc wrote:
> > On Tue, Mar 24, 2026 at 4:37 PM Richard Biener
> > <[email protected]> wrote:
> >> On Tue, Mar 24, 2026 at 9:25 AM Dongsheng Song <[email protected]>
> >> wrote:
> >>> Hi,
> >>>
> >>> I downloaded the archives of the most recent seven GCC test-result
> >>> emails. The results of the AI analysis are presented below for your
> >>> reference:
> [ ... ]
> I'd be careful drawing conclusions here.   It's interesting data, but I
> wouldn't necessarily equate it to port viability, popularity, or anything like
> that.   It's just measuring who's set up their CI bots to post to
> gcc-testresults.  If I look at build #s in my system (which would correspond
> to how many messages it would have sent to gcc-testresults if I let it) it
> would show that riscv is about 2X as popular as x86_64 which is just silly. 
> Even something like pru-elf would be seen as more popular than x86_64 because
> I often run the crosses for sniff testing without the natives.  Things like
> ppc would be at the bottom of the list because they're qemu emulated and with
> a 24+hr cycle time I only run them once a week).
> 
> A much more useful metric at least on the ISA side would be changes to the
> relevant config/ directory.  That won't get you granularity at the target
> level since a given ISA might have several ports for different OS variants. 
> THat's still going to be noisy data for a multitude of reasons, but at least
> gets you closer to measuring port popularity at the developer level.

I think the useful thing we get out of this data is what ports _not_
to consider (those w/o any coverage).  The data of course lacks any
info on the actual port status (number of FAIL, number of UNSUPPORTED)
and how those evolve over time.

Richard.

Reply via email to