Re: Status for armv7 vs. rust (and go)?

On May 24, 2023, at 16:08, Robert Clausecker <[email protected]> wrote:

> Hi Mark,

Hello.

> Am Wed, May 24, 2023 at 12:52:26PM -0700 schrieb Mark Millard:
>> May be things build but there are operational problems?
>> 
>> Anyway, I'd be interested to learn of whatever types of
>> armv7 problems rust may be currently having. (Rust being
>> a large build is a known issue.) 
> 
> The main problem with Rust ports on armv7 is that many of them have
> LLVM run into address space exhaustions, probably due to trying to
> build programs with LTO.  It's hard to diagnose these issues.
> 
> There are also a number of recurring portability problems due to
> time_t not being a long on armv7, but these are easy to patch.

I had forgotten, but on the HoneyComb (16 Cortex-A72 cores) I
have in its /usr/local/etc/poudriere.d/main-CA7-make.conf :

. . .
.if ${.CURDIR:M*/lang/*rust*}
MAKE_JOBS_NUMBER=8
.elif . . .
. . .

If I remember right, too many parallel jobs lead to hitting
a process size limit that stopped the build. I've not
experimented with between 8 and 16 (non-inclusive). With 8
I just had 1.70 build:

[00:00:14] [01] [00:00:00] Building lang/rust | rust-1.70.0
[02:16:24] [01] [02:16:10] Finished lang/rust | rust-1.70.0: Success

(The process size limit can be smaller for cortext-a72's doing
aarch32/armv7 activity than on a cortex-a7 armv7 system. That
might contribute  to hitting the issue.)

===
Mark Millard
marklmi at yahoo.com