Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-30 Thread Jonathan Yong via Gcc

On 3/24/26 08:36, LIU Hao wrote:

? 2026-3-24 10:52, Andrew Pinski via Gcc ??:

Also `i686-mingw32` should be swapped out with `x86_64-mingw`. I don't
think 32bit mingw is being testing these days only 64bit.


I would like to suggest adding `x86_64-w64-mingw32`, while changing 
`i686-mingw32` to `i686-w64-mingw32`.

`i686-mingw32` references the old mingw.org. They have lost control of their 
domain name, and osdn.net
which hosts their project has been down since months ago.

Please keep `i686-w64-mingw32`.

As both `{i686,x86_64}-pc-linux-gnu` are primary platforms, I don't see why 
`i686-w64-mingw32` can't be a
secondary platform. Almost all distributions of mingw-w64 provide both i686 and 
x86-64 builds: MSYS2,
Debian, ArchLinux, Ubuntu, mingw-builds, winlibs, and w64devkit. And I also 
build i686 myself. It won't
go away very soon.




Agreed, please keep i686-w64-mingw32 in secondary, the "w64" vendor is 
used to identify mingw-w64 based targets. It is mostly used to produce 
32-bit programs for Windows even if gcc itself is not running on 32-bit 
Windows.



Also I propose removing `i686-pc-cygwin`, I have not seen any recent
work on 32bit x86 cygwin or testresults.


I suggest swapping `x86_64-pc-cygwin` in.

Cygwin 3.4+ since 2022-11 is x86-64 only. As a downstream fork of Cygwin, MSYS2 
is also x86-64 only. Both
are key infrastructure for building GNU software on Windows, including GCC 
itself.




This is correct, 32bit i686-pc-cygwin is no longer an active project and 
only available through historical archives.




Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-26 Thread Andrew Pinski via Gcc
On Tue, Mar 24, 2026 at 6:21 AM Jason Merrill  wrote:
>
> Side note:  Please don't CC gcc@ and gcc-sc@ on the same message.

Oh, sorry about that. I updated the SC wiki
(https://gcc.gnu.org/wiki/gcc-steering-committee) to add a note for
others so they don't repeat my mistake.
The wording I used is `Note it is best not to CC the SC if you writing
an email to the GCC list.`

Thanks,
Andrew

>
> On Tue, Mar 24, 2026 at 2:54 AM Andrew Pinski via Gcc  wrote:
>>
>> Hi all,
>>   I would like to request some changes to the secondary tasrgets of
>> the GCC 17 Release Criteria. As the secondary targets seem out of date
>> in some cases.
>>
>> I propose to switch out mips64-linux-gnu with riscv64-linux-gnu as
>> being a secondary platform. riscv64-linux-gnu is gaining in popularity
>> and is riscv backend is actively maintained. There is much less
>> hardware available for mips64 too. Note we should maybe even consider
>> riscv64-linux-gnu for primary with GCC 18 but that might be too soon
>> at this point.
>>
>> Also `i686-mingw32` should be swapped out with `x86_64-mingw`. I don't
>> think 32bit mingw is being testing these days only 64bit.
>> Also I propose removing `i686-pc-cygwin`, I have not seen any recent
>> work on 32bit x86 cygwin or testresults.
>> Note I think adding `aarch64-mingw` would be too soon to add at this
>> stage. Though I think we should reconsider it for GCC 18.
>>
>> So the full for the secondary targets which I am proposing:
>> aarch64-elf
>> i686-apple-darwin
>> x86_64-mingw32
>> s390x-linux-gnu
>> riscv64-linux-gnu
>>
>>
>> Thanks,
>> Andrew
>>


Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-25 Thread Rainer Orth via Gcc
Hi Janne,

> On Tue, Mar 24, 2026 at 11:12 AM Rainer Orth via Gcc  wrote:
>>
>> Hi Dongsheng,
>>
>> thanks a lot for doing this.
>>
>> > ### Full Breakdown by Multiarch and GCC Major Version
>> >
>> > | Rank | Multiarch | Total | GCC 12 | GCC 13 | GCC 14 | GCC 15 | GCC 16 |
>> > |--|---|---||||||
>> [...]
>> > | 8 | i686-solaris | 258 | 0 | 0 | 58 | 58 | 142 |
>> > | 9 | sparc-solaris | 258 | 0 | 0 | 58 | 58 | 142 |
>> [...]
>> > | 15 | x86_64-solaris | 67 | 0 | 0 | 0 | 0 | 67 |
>> > | 16 | sparcv9-solaris | 67 | 0 | 0 | 0 | 0 | 67 |
>>
>> Given those numbers and the fact that sparc-sun-solaris2.11 is already a
>> primary platform, I wonder if we should make i386-pc-solaris2.11
>> secondary.  Sometimes there are significant differences between SPARC
>> and x86 here (often in favor of x86, actually).
>
> My 2c as a bystander these days: If you're planning to add another
> solaris target as a secondary platform, please pick x86_64, as AFAIU
> none of the upstreams (Oracle or Illumos distros like OpenIndiana)
> support anything except sparc and x86_64.

Solaris is always bi-arch, both x86 and sparc, and so would be this.
I've often found that 32-bit-default builds catch errors that are
missing in 64-bit-default ones.

Rainer

-- 
-
Rainer Orth, Center for Biotechnology, Bielefeld University


Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-25 Thread Janne Blomqvist via Gcc
On Tue, Mar 24, 2026 at 11:12 AM Rainer Orth via Gcc  wrote:
>
> Hi Dongsheng,
>
> thanks a lot for doing this.
>
> > ### Full Breakdown by Multiarch and GCC Major Version
> >
> > | Rank | Multiarch | Total | GCC 12 | GCC 13 | GCC 14 | GCC 15 | GCC 16 |
> > |--|---|---||||||
> [...]
> > | 8 | i686-solaris | 258 | 0 | 0 | 58 | 58 | 142 |
> > | 9 | sparc-solaris | 258 | 0 | 0 | 58 | 58 | 142 |
> [...]
> > | 15 | x86_64-solaris | 67 | 0 | 0 | 0 | 0 | 67 |
> > | 16 | sparcv9-solaris | 67 | 0 | 0 | 0 | 0 | 67 |
>
> Given those numbers and the fact that sparc-sun-solaris2.11 is already a
> primary platform, I wonder if we should make i386-pc-solaris2.11
> secondary.  Sometimes there are significant differences between SPARC
> and x86 here (often in favor of x86, actually).

My 2c as a bystander these days: If you're planning to add another
solaris target as a secondary platform, please pick x86_64, as AFAIU
none of the upstreams (Oracle or Illumos distros like OpenIndiana)
support anything except sparc and x86_64.

-- 
Janne Blomqvist


Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-25 Thread Mark Wielaard
Hi Jeff,

On Tue, Mar 24, 2026 at 07:46:24AM -0600, Jeffrey Law via Gcc wrote:
> On 3/24/2026 3:02 AM, Dongsheng Song via Gcc wrote:
> >On Tue, Mar 24, 2026 at 4:37 PM Richard Biener
> > wrote:
> >>On Tue, Mar 24, 2026 at 9:25 AM Dongsheng Song  
> >>wrote:
> >>>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.

To add a little more statistics (also arbitrary of course) here is the
prelimenary result of the Sourceware Survey 2026
https://sourceware.org/survey-2026 (it runs for another 9 days) for
the "builder.sourceware.org hardware?" question "Which architectures
have the highest priority, or which are missing?":

26%: x86_64
26%: arm64
10%: riscv64
 8%: i686
 5%: armhf
 5%: ppc64le
 5%: s390x
 3%: ppc64 

This is based on 40 responses, where more than half didn't answer he
question. So not sure how relevant it is.

I am surprised that in this survery arm64 is as important as x86_64.
riscv64 is more important than the 32bit i686 and armhf. And the big
endian arches come last.

We'll report the final tally when the survey ends (end of next week).
If you haven't taken the Survey yes, please do. It really helps the
Sourceware Project Leadership Committee decide what our infrastructure
priorities should be. https://sourceware.org/survey-2026

Thanks,

Mark


Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-24 Thread Jeffrey Law via Gcc




On 3/24/2026 7:58 AM, Richard Biener wrote:

.

Yes, it seems to be still fastest to do qemu-user based testing on
x86-64 for riscv ...
The K1 and my 40c skylake systems were pretty comparable.  The former 
generates a lot less heat and doesn't compete with other builds though.  
  I'd expect QEMU to win when hosted on something more modern than 
skylake cores ;-)




But I guess riscv(32?) is relevant as a cross target for embedded use.
Perhaps.  There's certainly a lot of them around, but I don't see many, 
if any, GCC focused focused on the rv32 space.  It obviously benefits 
from the working going on for rv64.


Jeff


Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-24 Thread Richard Biener via Gcc
On Tue, 24 Mar 2026, Jeffrey Law wrote:

> 
> 
> On 3/24/2026 2:03 AM, Richard Biener via Gcc wrote:
> > I'd like to see comments from the respective old/new triplets
> > target/OS maintainers,
> > in particular this request should probably be split up.
> >
> > I agree to _add_ riscv64-linux-gnu for GCC 17.
> That seems quite reasonable and I wouldn't go any further than that for riscv
> until such time as there's dramatically better performing hardware with wide
> availability.   And no, the upcoming K3s don't meet my view of dramatically
> better.
> 
> My x86_64 tests run in 1-2hrs on skylake era systems.  riscv on the K1 designs
> is nearly 30 hours.  At best I expect the K3  to cut that down to 15 hours.   
> The pioneer (c920 design) is just 7-8 hours, but it's been discontinued and
> they've been pretty flakey.

Yes, it seems to be still fastest to do qemu-user based testing on 
x86-64 for riscv ...

But I guess riscv(32?) is relevant as a cross target for embedded use.

This is a list of secondary _targets_, not hosts, after all ;)

Richard.


Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-24 Thread Jeffrey Law via Gcc




On 3/24/2026 7:50 AM, Richard Biener wrote:

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.
Yea, that's probably a reasonable way to look at it.  If nothing's going 
to that list, then I woudn't consider it at all.


Jeff


Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-24 Thread Jeffrey Law via Gcc




On 3/24/2026 2:03 AM, Richard Biener via Gcc wrote:

I'd like to see comments from the respective old/new triplets
target/OS maintainers,
in particular this request should probably be split up.

I agree to _add_ riscv64-linux-gnu for GCC 17.
That seems quite reasonable and I wouldn't go any further than that for 
riscv until such time as there's dramatically better performing hardware 
with wide availability.   And no, the upcoming K3s don't meet my view of 
dramatically better.


My x86_64 tests run in 1-2hrs on skylake era systems.  riscv on the K1 
designs is nearly 30 hours.  At best I expect the K3  to cut that down 
to 15 hours.    The pioneer (c920 design) is just 7-8 hours, but it's 
been discontinued and they've been pretty flakey.


Jeff


Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-24 Thread Richard Biener via Gcc
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
> >  wrote:
> >> On Tue, Mar 24, 2026 at 9:25 AM Dongsheng Song 
> >> 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.


Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-24 Thread Jeffrey Law via Gcc




On 3/24/2026 3:02 AM, Dongsheng Song via Gcc wrote:

On Tue, Mar 24, 2026 at 4:37 PM Richard Biener
 wrote:

On Tue, Mar 24, 2026 at 9:25 AM Dongsheng Song  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.


Jeff


Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-24 Thread Jason Merrill via Gcc
Side note:  Please don't CC gcc@ and gcc-sc@ on the same message.

On Tue, Mar 24, 2026 at 2:54 AM Andrew Pinski via Gcc 
wrote:

> Hi all,
>   I would like to request some changes to the secondary tasrgets of
> the GCC 17 Release Criteria. As the secondary targets seem out of date
> in some cases.
>
> I propose to switch out mips64-linux-gnu with riscv64-linux-gnu as
> being a secondary platform. riscv64-linux-gnu is gaining in popularity
> and is riscv backend is actively maintained. There is much less
> hardware available for mips64 too. Note we should maybe even consider
> riscv64-linux-gnu for primary with GCC 18 but that might be too soon
> at this point.
>
> Also `i686-mingw32` should be swapped out with `x86_64-mingw`. I don't
> think 32bit mingw is being testing these days only 64bit.
> Also I propose removing `i686-pc-cygwin`, I have not seen any recent
> work on 32bit x86 cygwin or testresults.
> Note I think adding `aarch64-mingw` would be too soon to add at this
> stage. Though I think we should reconsider it for GCC 18.
>
> So the full for the secondary targets which I am proposing:
> aarch64-elf
> i686-apple-darwin
> x86_64-mingw32
> s390x-linux-gnu
> riscv64-linux-gnu
>
>
> Thanks,
> Andrew
>
>


Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-24 Thread Rainer Orth via Gcc
Hi Dongsheng,

thanks a lot for doing this.

> ### Full Breakdown by Multiarch and GCC Major Version
>
> | Rank | Multiarch | Total | GCC 12 | GCC 13 | GCC 14 | GCC 15 | GCC 16 |
> |--|---|---||||||
[...]
> | 8 | i686-solaris | 258 | 0 | 0 | 58 | 58 | 142 |
> | 9 | sparc-solaris | 258 | 0 | 0 | 58 | 58 | 142 |
[...]
> | 15 | x86_64-solaris | 67 | 0 | 0 | 0 | 0 | 67 |
> | 16 | sparcv9-solaris | 67 | 0 | 0 | 0 | 0 | 67 |

Given those numbers and the fact that sparc-sun-solaris2.11 is already a
primary platform, I wonder if we should make i386-pc-solaris2.11
secondary.  Sometimes there are significant differences between SPARC
and x86 here (often in favor of x86, actually).

Rainer

-- 
-
Rainer Orth, Center for Biotechnology, Bielefeld University


Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-24 Thread Dongsheng Song via Gcc
On Tue, Mar 24, 2026 at 4:37 PM Richard Biener
 wrote:
>
> On Tue, Mar 24, 2026 at 9:25 AM Dongsheng Song  
> 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:
> >
> > Key Points:
> >
> > - powerpc64le and powerpc64 combined account for 67% of the total;
> > they are clearly the dominant platforms, evidently supported by a
> > highly active fleet of automated testing bots.
> > - x86_64 spans four operating systems (Linux, Darwin, FreeBSD, and
> > Solaris) and accounts for approximately 16% of the total.
> > - aarch64-linux-gnu accounts for only 3.5%; considering the widespread
> > adoption of ARM servers, this level of test coverage is notably low.
> > - riscv64-linux-gnu accounts for a mere 38 reports (0.3%), while
> > loongarch64-linux-gnu accounts for only 5 reports; test coverage for
> > these emerging architectures is severely inadequate.
> > - Traditional or niche architectures—such as ia64, mmix, mips64el, and
> > vax—each account for only 1 to 3 reports, indicating that they are
> > currently receiving only minimal maintenance support.
> >
> > The statistical results are presented below (Data scope: September
> > 2025 through March 2026—a 7-month period—comprising 15,151 test
> > reports across 29 distinct Multiarch platforms):
>
> That's interesting - did you manually filter the GCC version tested to
> a single one?  Otherwise the result isn't telling much, at least for the
> entries with very few reports, but maybe also in general.
>

Here are the statistical analysis results, including the GCC major
version numbers:

## GCC Test Results Mailing List Analysis

Data source: gcc-testresults mailing list archives, 2025-09 ~ 2026-03
(7 months), 15,151 test reports across 29 multiarch
targets.

### Key Findings

- GCC 16 (development trunk) accounts for 58.1% of all reports — most
test bots track the master branch.
- powerpc64le + powerpc64 dominate with 67% of all reports, clearly
driven by very active automated test bots. They are the
only architecture with continuous testing across all GCC versions
(12/13/14/15/16).
- x86_64 spans 4 OSes (Linux/Darwin/FreeBSD/Solaris) totaling ~16%,
but given its market dominance, this is surprisingly
low.
- aarch64-linux-gnu has only 3.5% coverage — underwhelming given the
rapid adoption of ARM servers.
- x86_64-darwin (684 reports) is almost entirely GCC 16, meaning macOS
testers only care about the latest development
version.
- GCC 12 has only 467 reports, mostly from powerpc64 — other
architectures have essentially stopped testing GCC 12.
- riscv64-linux-gnu has just 38 reports (0.3%) with zero coverage for
GCC 14, showing a significant gap.
- loongarch64-linux-gnu has only 5 reports total — severely undertested.
- avr-elf is one of the few embedded targets with relatively even
coverage across GCC 13/14/15/16.
- Legacy/niche architectures like ia64, mmix, mips64el, and vax have
1–3 reports each, at bare minimum maintenance level.

### Full Breakdown by Multiarch and GCC Major Version

| Rank | Multiarch | Total | GCC 12 | GCC 13 | GCC 14 | GCC 15 | GCC 16 |
|--|---|---||||||
| 1 | powerpc64le-linux-gnu | 6,146 | 2 | 811 | 1,082 | 1,115 | 3,135 |
| 2 | powerpc64-linux-gnu | 4,005 | 437 | 710 | 760 | 766 | 1,332 |
| 3 | x86_64-linux-gnu | 1,449 | 2 | 2 | 60 | 121 | 1,209 |
| 4 | x86_64-darwin | 684 | 0 | 0 | 0 | 1 | 683 |
| 5 | aarch64-linux-gnu | 526 | 5 | 0 | 0 | 11 | 496 |
| 6 | s390x-linux-gnu | 429 | 5 | 3 | 5 | 4 | 412 |
| 7 | arm-eabi | 260 | 0 | 0 | 0 | 0 | 260 |
| 8 | i686-solaris | 258 | 0 | 0 | 58 | 58 | 142 |
| 9 | sparc-solaris | 258 | 0 | 0 | 58 | 58 | 142 |
| 10 | pru-elf | 218 | 0 | 6 | 6 | 13 | 193 |
| 11 | x86_64-freebsd | 208 | 0 | 0 | 29 | 0 | 179 |
| 12 | i686-darwin | 146 | 0 | 0 | 0 | 1 | 145 |
| 13 | m68k-linux-gnu | 132 | 0 | 0 | 0 | 0 | 132 |
| 14 | arm-linux-gnueabihf | 116 | 5 | 4 | 5 | 4 | 98 |
| 15 | x86_64-solaris | 67 | 0 | 0 | 0 | 0 | 67 |
| 16 | sparcv9-solaris | 67 | 0 | 0 | 0 | 0 | 67 |
| 17 | avr-elf | 48 | 0 | 11 | 12 | 12 | 13 |
| 18 | riscv64-linux-gnu | 38 | 3 | 1 | 0 | 3 | 31 |
| 19 | hppa64-hpux | 29 | 0 | 0 | 0 | 0 | 29 |
| 20 | hppa-linux-gnu | 17 | 0 | 0 | 0 | 0 | 17 |
| 21 | i686-linux-gnu | 16 | 5 | 3 | 5 | 0 | 3 |
| 22 | arm-linux-gnueabi | 10 | 3 | 1 | 2 | 2 | 2 |
| 23 | powerpc-aix | 8 | 0 | 0 | 0 | 0 | 8 |
| 24 | loongarch64-linux-gnu | 5 | 0 | 0 | 0 | 4 | 1 |
| 25 | riscv32-elf | 5 | 0 | 0 | 0 | 2 | 3 |
| 26 | ia64-linux-gnu | 3 | 0 | 0 | 0 | 0 | 3 |
| 27 | mmix-mmixware | 1 | 0 | 0 | 0 | 0 | 1 |
| 28 | mips64el-linux-gnuabi64 | 1 | 0 | 0 | 0 | 0 | 1 |
| 29 | vax-netbsdelf | 1 | 0 | 0 | 0 | 0 | 1 |
| | Total | 15,151 | 467 | 1,552 | 2,082 | 2,175 | 8,805 |

Thanks,
Dongsheng


GCC-test-results-stat-2026034-vers.md
Description: Binary data


Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-24 Thread Richard Biener via Gcc
On Tue, Mar 24, 2026 at 9:25 AM Dongsheng Song  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:
>
> Key Points:
>
> - powerpc64le and powerpc64 combined account for 67% of the total;
> they are clearly the dominant platforms, evidently supported by a
> highly active fleet of automated testing bots.
> - x86_64 spans four operating systems (Linux, Darwin, FreeBSD, and
> Solaris) and accounts for approximately 16% of the total.
> - aarch64-linux-gnu accounts for only 3.5%; considering the widespread
> adoption of ARM servers, this level of test coverage is notably low.
> - riscv64-linux-gnu accounts for a mere 38 reports (0.3%), while
> loongarch64-linux-gnu accounts for only 5 reports; test coverage for
> these emerging architectures is severely inadequate.
> - Traditional or niche architectures—such as ia64, mmix, mips64el, and
> vax—each account for only 1 to 3 reports, indicating that they are
> currently receiving only minimal maintenance support.
>
> The statistical results are presented below (Data scope: September
> 2025 through March 2026—a 7-month period—comprising 15,151 test
> reports across 29 distinct Multiarch platforms):

That's interesting - did you manually filter the GCC version tested to
a single one?  Otherwise the result isn't telling much, at least for the
entries with very few reports, but maybe also in general.


> | Rank | Report Count | Percentage | Multiarch |
> |--||--|---|
> | 1 | 6,146 | 40.6% | powerpc64le-linux-gnu |
> | 2 | 4,005 | 26.4% | powerpc64-linux-gnu |
> | 3 | 1,449 | 9.6% | x86_64-linux-gnu |
> | 4 | 684 | 4.5% | x86_64-darwin |
> | 5 | 526 | 3.5% | aarch64-linux-gnu |
> | 6 | 429 | 2.8% | s390x-linux-gnu |
> | 7 | 260 | 1.7% | arm-eabi |
> | 8 | 258 | 1.7% | i686-solaris |
> | 9 | 258 | 1.7% | sparc-solaris |
> | 10 | 218 | 1.4% | pru-elf |
> | 11 | 208 | 1.4% | x86_64-freebsd |
> | 12 | 146 | 1.0% | i686-darwin |
> | 13 | 132 | 0.9% | m68k-linux-gnu |
> | 14 | 116 | 0.8% | arm-linux-gnueabihf |
> | 15 | 67 | 0.4% | x86_64-solaris |
> | 16 | 67 | 0.4% | sparcv9-solaris |
> | 17 | 48 | 0.3% | avr-none |
> | 18 | 38 | 0.3% | riscv64-linux-gnu |
> | 19 | 29 | 0.2% | hppa64-hpux |
> | 20 | 17 | 0.1% | hppa-linux-gnu |
> | 21 | 16 | 0.1% | i686-linux-gnu |
> | 22 | 10 | 0.1% | arm-linux-gnueabi |
> | 23 | 8 | 0.1% | powerpc-aix |
> | 24 | 5 | 0.0% | loongarch64-linux-gnu |
> | 25 | 5 | 0.0% | riscv32-none-elf |
> | 26 | 3 | 0.0% | ia64-linux-gnu |
> | 27 | 1 | 0.0% | mmix-mmixware |
> | 28 | 1 | 0.0% | mips64el-linux-gnuabi64 |
> | 29 | 1 | 0.0% | vax-netbsdelf |
>
> Thanks,
> Dongsheng
>
>
> On Tue, Mar 24, 2026 at 4:05 PM Richard Biener via Gcc  
> wrote:
> >
> > On Tue, Mar 24, 2026 at 3:54 AM Andrew Pinski via Gcc  
> > wrote:
> > >
> > > Hi all,
> > >   I would like to request some changes to the secondary tasrgets of
> > > the GCC 17 Release Criteria. As the secondary targets seem out of date
> > > in some cases.
> > >
> > > I propose to switch out mips64-linux-gnu with riscv64-linux-gnu as
> > > being a secondary platform. riscv64-linux-gnu is gaining in popularity
> > > and is riscv backend is actively maintained. There is much less
> > > hardware available for mips64 too. Note we should maybe even consider
> > > riscv64-linux-gnu for primary with GCC 18 but that might be too soon
> > > at this point.
> > >
> > > Also `i686-mingw32` should be swapped out with `x86_64-mingw`. I don't
> > > think 32bit mingw is being testing these days only 64bit.
> > > Also I propose removing `i686-pc-cygwin`, I have not seen any recent
> > > work on 32bit x86 cygwin or testresults.
> > > Note I think adding `aarch64-mingw` would be too soon to add at this
> > > stage. Though I think we should reconsider it for GCC 18.
> > >
> > > So the full for the secondary targets which I am proposing:
> > > aarch64-elf
> > > i686-apple-darwin
> > > x86_64-mingw32
> > > s390x-linux-gnu
> > > riscv64-linux-gnu
> >
> > I'd like to see comments from the respective old/new triplets
> > target/OS maintainers,
> > in particular this request should probably be split up.
> >
> > I agree to _add_ riscv64-linux-gnu for GCC 17.
> >
> > Thanks,
> > Richard.
> >
> > >
> > >
> > > Thanks,
> > > Andrew


Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-24 Thread LIU Hao via Gcc

在 2026-3-24 10:52, Andrew Pinski via Gcc 写道:

Also `i686-mingw32` should be swapped out with `x86_64-mingw`. I don't
think 32bit mingw is being testing these days only 64bit.


I would like to suggest adding `x86_64-w64-mingw32`, while changing 
`i686-mingw32` to `i686-w64-mingw32`.

`i686-mingw32` references the old mingw.org. They have lost control of their domain name, and osdn.net 
which hosts their project has been down since months ago.


Please keep `i686-w64-mingw32`.

As both `{i686,x86_64}-pc-linux-gnu` are primary platforms, I don't see why `i686-w64-mingw32` can't be a 
secondary platform. Almost all distributions of mingw-w64 provide both i686 and x86-64 builds: MSYS2, 
Debian, ArchLinux, Ubuntu, mingw-builds, winlibs, and w64devkit. And I also build i686 myself. It won't 
go away very soon.




Also I propose removing `i686-pc-cygwin`, I have not seen any recent
work on 32bit x86 cygwin or testresults.


I suggest swapping `x86_64-pc-cygwin` in.

Cygwin 3.4+ since 2022-11 is x86-64 only. As a downstream fork of Cygwin, MSYS2 is also x86-64 only. Both 
are key infrastructure for building GNU software on Windows, including GCC itself.




Note I think adding `aarch64-mingw` would be too soon to add at this
stage. Though I think we should reconsider it for GCC 18.


Yes.



--
Best regards,
LIU Hao


OpenPGP_signature.asc
Description: OpenPGP digital signature


Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-24 Thread Dongsheng Song via Gcc
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:

Key Points:

- powerpc64le and powerpc64 combined account for 67% of the total;
they are clearly the dominant platforms, evidently supported by a
highly active fleet of automated testing bots.
- x86_64 spans four operating systems (Linux, Darwin, FreeBSD, and
Solaris) and accounts for approximately 16% of the total.
- aarch64-linux-gnu accounts for only 3.5%; considering the widespread
adoption of ARM servers, this level of test coverage is notably low.
- riscv64-linux-gnu accounts for a mere 38 reports (0.3%), while
loongarch64-linux-gnu accounts for only 5 reports; test coverage for
these emerging architectures is severely inadequate.
- Traditional or niche architectures—such as ia64, mmix, mips64el, and
vax—each account for only 1 to 3 reports, indicating that they are
currently receiving only minimal maintenance support.

The statistical results are presented below (Data scope: September
2025 through March 2026—a 7-month period—comprising 15,151 test
reports across 29 distinct Multiarch platforms):

| Rank | Report Count | Percentage | Multiarch |
|--||--|---|
| 1 | 6,146 | 40.6% | powerpc64le-linux-gnu |
| 2 | 4,005 | 26.4% | powerpc64-linux-gnu |
| 3 | 1,449 | 9.6% | x86_64-linux-gnu |
| 4 | 684 | 4.5% | x86_64-darwin |
| 5 | 526 | 3.5% | aarch64-linux-gnu |
| 6 | 429 | 2.8% | s390x-linux-gnu |
| 7 | 260 | 1.7% | arm-eabi |
| 8 | 258 | 1.7% | i686-solaris |
| 9 | 258 | 1.7% | sparc-solaris |
| 10 | 218 | 1.4% | pru-elf |
| 11 | 208 | 1.4% | x86_64-freebsd |
| 12 | 146 | 1.0% | i686-darwin |
| 13 | 132 | 0.9% | m68k-linux-gnu |
| 14 | 116 | 0.8% | arm-linux-gnueabihf |
| 15 | 67 | 0.4% | x86_64-solaris |
| 16 | 67 | 0.4% | sparcv9-solaris |
| 17 | 48 | 0.3% | avr-none |
| 18 | 38 | 0.3% | riscv64-linux-gnu |
| 19 | 29 | 0.2% | hppa64-hpux |
| 20 | 17 | 0.1% | hppa-linux-gnu |
| 21 | 16 | 0.1% | i686-linux-gnu |
| 22 | 10 | 0.1% | arm-linux-gnueabi |
| 23 | 8 | 0.1% | powerpc-aix |
| 24 | 5 | 0.0% | loongarch64-linux-gnu |
| 25 | 5 | 0.0% | riscv32-none-elf |
| 26 | 3 | 0.0% | ia64-linux-gnu |
| 27 | 1 | 0.0% | mmix-mmixware |
| 28 | 1 | 0.0% | mips64el-linux-gnuabi64 |
| 29 | 1 | 0.0% | vax-netbsdelf |

Thanks,
Dongsheng


On Tue, Mar 24, 2026 at 4:05 PM Richard Biener via Gcc  wrote:
>
> On Tue, Mar 24, 2026 at 3:54 AM Andrew Pinski via Gcc  wrote:
> >
> > Hi all,
> >   I would like to request some changes to the secondary tasrgets of
> > the GCC 17 Release Criteria. As the secondary targets seem out of date
> > in some cases.
> >
> > I propose to switch out mips64-linux-gnu with riscv64-linux-gnu as
> > being a secondary platform. riscv64-linux-gnu is gaining in popularity
> > and is riscv backend is actively maintained. There is much less
> > hardware available for mips64 too. Note we should maybe even consider
> > riscv64-linux-gnu for primary with GCC 18 but that might be too soon
> > at this point.
> >
> > Also `i686-mingw32` should be swapped out with `x86_64-mingw`. I don't
> > think 32bit mingw is being testing these days only 64bit.
> > Also I propose removing `i686-pc-cygwin`, I have not seen any recent
> > work on 32bit x86 cygwin or testresults.
> > Note I think adding `aarch64-mingw` would be too soon to add at this
> > stage. Though I think we should reconsider it for GCC 18.
> >
> > So the full for the secondary targets which I am proposing:
> > aarch64-elf
> > i686-apple-darwin
> > x86_64-mingw32
> > s390x-linux-gnu
> > riscv64-linux-gnu
>
> I'd like to see comments from the respective old/new triplets
> target/OS maintainers,
> in particular this request should probably be split up.
>
> I agree to _add_ riscv64-linux-gnu for GCC 17.
>
> Thanks,
> Richard.
>
> >
> >
> > Thanks,
> > Andrew


Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-24 Thread Richard Biener via Gcc
On Tue, Mar 24, 2026 at 3:54 AM Andrew Pinski via Gcc  wrote:
>
> Hi all,
>   I would like to request some changes to the secondary tasrgets of
> the GCC 17 Release Criteria. As the secondary targets seem out of date
> in some cases.
>
> I propose to switch out mips64-linux-gnu with riscv64-linux-gnu as
> being a secondary platform. riscv64-linux-gnu is gaining in popularity
> and is riscv backend is actively maintained. There is much less
> hardware available for mips64 too. Note we should maybe even consider
> riscv64-linux-gnu for primary with GCC 18 but that might be too soon
> at this point.
>
> Also `i686-mingw32` should be swapped out with `x86_64-mingw`. I don't
> think 32bit mingw is being testing these days only 64bit.
> Also I propose removing `i686-pc-cygwin`, I have not seen any recent
> work on 32bit x86 cygwin or testresults.
> Note I think adding `aarch64-mingw` would be too soon to add at this
> stage. Though I think we should reconsider it for GCC 18.
>
> So the full for the secondary targets which I am proposing:
> aarch64-elf
> i686-apple-darwin
> x86_64-mingw32
> s390x-linux-gnu
> riscv64-linux-gnu

I'd like to see comments from the respective old/new triplets
target/OS maintainers,
in particular this request should probably be split up.

I agree to _add_ riscv64-linux-gnu for GCC 17.

Thanks,
Richard.

>
>
> Thanks,
> Andrew


Re: changes requesting for GCC 17 Release Criteria (secondary targets)

2026-03-24 Thread Iain Sandoe via Gcc



> On 24 Mar 2026, at 02:52, Andrew Pinski via Gcc  wrote:
> 
> Hi all,
>  I would like to request some changes to the secondary tasrgets of
> the GCC 17 Release Criteria. As the secondary targets seem out of date
> in some cases.
> 
> I propose to switch out mips64-linux-gnu with riscv64-linux-gnu as
> being a secondary platform. riscv64-linux-gnu is gaining in popularity
> and is riscv backend is actively maintained. There is much less
> hardware available for mips64 too. Note we should maybe even consider
> riscv64-linux-gnu for primary with GCC 18 but that might be too soon
> at this point.
> 
> Also `i686-mingw32` should be swapped out with `x86_64-mingw`. I don't
> think 32bit mingw is being testing these days only 64bit.
> Also I propose removing `i686-pc-cygwin`, I have not seen any recent
> work on 32bit x86 cygwin or testresults.
> Note I think adding `aarch64-mingw` would be too soon to add at this
> stage. Though I think we should reconsider it for GCC 18.
> 
> So the full for the secondary targets which I am proposing:
> aarch64-elf
> i686-apple-darwin

This probably should be x86_64-apple-darwin, actually I have a dim memory
that we (collectively) might have had that discussion before.  i686-darwin is
still tested, but very much less than x86_64.

thanks
Iain

> x86_64-mingw32
> s390x-linux-gnu
> riscv64-linux-gnu
> 
> 
> Thanks,
> Andrew