Re: CVS commit: pkgsrc/graphics/librsvg

[Havard Eidnes <he%NetBSD.org@localhost>]

>> Module Name:	pkgsrc
>> Committed By:	he
>> Date:		Fri Oct 29 06:55:10 UTC 2021
>>
>> Modified Files:
>> 	pkgsrc/graphics/librsvg: available.mk
>>
>> Log Message:
>> Remove the special-case disabling of earm and powerpc,
>> which references my own old post from mid-2020 at which time
>> I had problems building rust for earmv7hf "natively".
>> That problem has long since been resolved.  Not sure why
>> powerpc got the same treatment -- with this removed this package
>> builds fine with rust 1.56.0 on NetBSD/macppc 9.99.88.
>
> Do you mean that if I take a typical earm machine like an RPI3 (1G RAM),
> and cehck out pkgsrc and do a "make package-install" in something that
> depends on librsvg, you think that it will build rust, build librsvg,
> and continue, without any issues?

Yes, modulo whether 1GB physical RAM is sufficient for it to
finish in finite time.

The limitation which prevented the earmhf7f build to complete and
which I mentioned in the now-removed link to the message was the
size of the virtual address space on this platform.  It is in the
area of 1.8GB (was bumped from 1.5GB, but it's apparently hard or
impossible to go above 1.8GB), and some of the build processes
used to exceed that limit.

Doing the build on an RPI3 with only 1G RAM may turn out to be an
excercise in patience (swap will be needed).

I run my builds on a qemu-running earmv7hf with enough "real"
RAM, but the user processes are still subject to the virtual
address space restriction of the 32-bit arm ports.  I have been
in the habit of uploading the binary packages which result from a
"make package" on this virtualized host (i.e. not "bootstrap
kits"), ref.

  ftp://ftp.NetBSD.org/pub/NetBSD/misc/he/

At one point the build could not be completed because the virtual
size was exceeded, but that hasn't been the case since version
1.44.0 by the look of the contents of the above directory.

> I have been under the impression that building rust required
> unreasoably large (by our standards) amounts of RAM, but I'll
> be very happy to hear that this has been fixed upstream.

While building rust itself certainly is resource intensive, I
don't see that as a reason to exclude the package from being able
to build at all.  It's not as if RPI3 with 1GB RAM is the
maximum-size host which can run the 32-bit arm port, even if you
discount qemu.  It's another matter if it can't be built at all
on a given architecture.

In my case, the librsvg package also completes the build on my
emulated earmv7hf system.  I cannot claim to have tested the
result or any package which uses that package; assistance on that
front is gratefully accepted.

(Also, the build error I got when that test was present was not
exactly obvious -- the build complained that harfbuzz needs
graphite2's pkgconfig file, while that package certainly is
installed, it's just not buildlink'ed caused by the LIBRSVG_TYPE
was being set to 'c'...)

Best regards,

- Håvard