[patch] Important comment update to gcc/configure.in
Nathanael Nerode
neroden@twcny.rr.com
Fri Mar 14 19:45:00 GMT 2003
Daniel Jacobowitz wrote:
> On Fri, Mar 14, 2003 at 02:34:14PM -0500, Nathanael Nerode wrote:
>
>>Daniel Jacobowitz wrote:
>>
>>>On Fri, Mar 14, 2003 at 12:09:36AM -0500, Nathanael Nerode wrote:
>>>
>>>
>>>>I believe this is correct. I want review only in case I got it wrong.
>>>>
>>>>This clarifies the important issue of *which* assembler and linker
>>>>we are looking for -- it's not the BUILD->BUILD, BUILD->HOST, or
>>>>BUILD->TARGET assemblers (all of which might be used and different in
>>>>a Canadian cross compile)... it's the HOST->TARGET one, which we can't
>>>>always even execute.
>>>>
>>>>Now that I've got this clear, I believe I'll actually be able to clean
>>>>up the related code.
>>>
>>>
>>>Nope. Couple of subtleties here, I think. First of all, remember,
>>>we're on BUILD right now. We can not search for anything that's
>>>supposed to live on HOST; it might not be there. Secondly, we use this
>>>assembler for feature tests, so we have to be able to execute it.
>>
>>Ah, but that's what the version checks are for. :-) For when we are
>>building a brand new HOST->TARGET gas and can't run it.
>
>
> That's not the interesting case. Just because BUILD != HOST doesn't
> imply we're building gas in-tree. I never am. I still need to get the
> features somehow.
>
>
>>>Generally we want a BUILD->TARGET assembler, and to make the assumption
>>>that it will have the same features as the HOST->TARGET assembler.
>>
>>That's really confusing. :-) That seriously needs to be documented,
>>since it's an undocumented assumption if BUILD!=HOST.
>>
>>The fact is that the feature tests are used as if they were feature
>>tests for the HOST->TARGET assembler. They determine the format of the
>>assembly output by the newly built gcc. I *think* that's the only thing
>>they determine, but I'm not sure yet.
>>
>>Because of this, in the build!=host case, the version number checks on
>>the newly built gas are actually *better*, more accurate tests, than the
>>feature checks, which are on a build->target assembler, which isn't the
>>assembler we're actually interested in. It's an interesting situation,
>>certainly...
>
>
> Except where they're impossible.
Of course. My point was that getting rid of the version number checks
is undesirable because they're actually the best way of getting
information in a particular setup. Ick, but true.
--Nathanael
More information about the Gcc-patches
mailing list