Re: Building on MinGW using MXE-built dependencies [WAS: Re: mxe-installer try 2]

[Philip Nienhuis]

Philip Nienhuis wrote:
> John W. Eaton wrote:
> > On 06/11/2013 03:43 PM, Philip Nienhuis wrote:
> > > John W. Eaton wrote:
> > > > On 06/11/2013 02:51 PM, Philip Nienhuis wrote:
> > > >
> > > > > Here I didn't even get as far as you did. I installed mingw + above
> > > > > dependencies, updated/upgraded, cloned mxe-octave.
> > > > > The mxe build breaks while building gcc.
> > > >
> > > > That's not currently expected to work. I've not been able to get gcc to
> > > > build for any native build setup. So you need to try with
> > > > USE_SYSTEM_GCC
> > > > set to yes.
> > >
> > > Stupid me, I could have seen that myself, didn't think of reading that
> > > far into Makefile. Sorry.
> > >
> > > Anyway, I changed USE_SYSTEM_GCC to "yes" but now ./mk-dist insists on
> > > building gcc. How can I avoid that?
> > >
> > > Philip
> >
> > Are you sure it is actually building GCC? It will still unpack and run
> > the build target, but if you look at gcc.mk, you'll see that when
> > USE_SYSTEM_GCC is set to yes, it will only create a cmake configuration
> > file, not actually build the compiler.
>
> Hmmm, so the message "[build] gcc" (and the long period of hard disk
> activity) is a bit deceptive....
>
> But I saw you are correct, it started building blas and now lapack.

...and that MinGW build (on WinXP 32b) ended in building postgres 
complaining about cpio, just like John D reported earlier. So that looks 
to be a consistent result. (BTW I changed the postgres configure option 
in postgres.mk into "--without-zlib" to get past the zlib error that 
J.D. reported as well)

Just out of curiosity: why are postgres and libodbc required anyway? 
Same for llvm, if it isn't going to be used for the next stable release 
(as you suggested)? - llvm isn't called in octave.mk (no --enable-jit 
configure option.) Together these builds make up for around 15 % of 
total build time, so it should help speeding up this first stage of 
getting the native MinGW to succeed if we could (temporarily) avoid 
building them. Especially as multi-core builds don't work on native MinGW.

(To John D:
Anti-virus (or better: anti-malware) SW interfering on Windows is a 
well-known issue; AVG is known to be one of the more paranoid variants. 
I use MS Essentials, it is less picky but at the same time perhaps less 
"secure". On Linux I have clamav installed but that doesn't do on-access 
scanning - if it did we might get similar issues on Linux.
On a related note, I had to disable UAC on Windows 7 to avoid random 
stops anywhere during MinGW builds)

Philip