[Rd] Trying to compile R 4.2.x on Linux as 32bit sub-architecture

[Tomas Kalibera]

On 6/20/22 15:40, Berwin A Turlach wrote:
> G'day Tomas,
>
> On Mon, 20 Jun 2022 14:46:04 +0200
> Tomas Kalibera <tomas.kalibera using gmail.com> wrote:
>
>>> Well, what puzzles me is that I had no problem with R-patched right
>>> until the end of the 4.1 branch.  And I have no problems with
>>> R-devel, that is also still installing a 32bit and a 64bit version
>>> just fine.
>>>
>>> It is just R 4.2 that gives me the problem.
>> Ah, that's interesting. And could you perhaps bisect it to the
>> problematic commit? That might increase the chances you get attention
>> of the person who authored it. After all you could also send a patch
> Well, I found via Google a command that could tell me the first revision
> of a branch.  Apparently revision 81976 is the first revision of the 4.2
> branch:
>
> ------------------------------------------------------------------------
> r81976 | pd | 2022-03-25 07:02:01 +0800 (Fri, 25 Mar 2022) | 1 line
>
> Create R-4-2-branch
> ------------------------------------------------------------------------
>
> I checked this branch out and it already displays the same problem,
> the same test in reg-tests-1d.R falls over.  But I am pretty sure that
> R-devel (and R-patched) compiled fine on that day. So where was the
> change made before the R-4.2-branch was created?

the R-4-2-branch was created from R-devel branch by that commit you 
found, so the previous changes were on R-devel. So you could now 
actually search for a commit on R-devel (after r81978) that fixed it, 
perhaps it is something that could be ported to R-4-2-branch.

You could restrict the search for changes in the corresponding test 
files, first, checking if the tests/the tolerances have changed (that 
would be easy, "svn blame" would help: "svn blame", "svn blame -r xxxx", 
"svn log --diff", etc). If it is not a change in the tests but rather in 
the computations, that would be harder to guess, still bisecting should 
find it.

Best
Tomas

>
> Cheers,
>
> 	Berwin