Re: [Monotone-devel] Mingw 64 bit build

[Stephen Leake]

Markus Wanner <address@hidden> writes:

> Stephen,
>
> On 05/03/2014 03:54 PM, Stephen Leake wrote:
>> I've built nvm.msys2-mingw-64 with Mys2/MingW64 64 bit. It passes all tests
>> except one func test (empty_environment) and some extra tests.
>
> Cool, thanks.
>
> What goes wrong in empty_environment? That one works on Msys 1.0.

The commit gives:

stderr:
mtn: beginning commit on branch 'testbranch'
mtn: error: sqlite error: SQL logic error or missing database

I'm guessing it's a library issue, combined with the weird way msys2
works. 

>> I need to test that branch on a Unix box (I have Debian and Cygwin),
>> since most of the changes are in '#ifdef Windows' areas.
>
> I just tried: There are a couple of places where Unix needs an argument
> that you've commented out. Please revert those.

Done in nvm.cleanup-warnings.

> Also, I'm not a fan of the cast to void hack. That clutters the source
> code quite a bit - especially when you need additional ifdefs to filter
> based on platform. I'd rather disable that warning.

I reviewed all the FIXME-UNUSED, and deleted some because they were not
needed. So -Wno-unused is useful.

There's not much clutter; I think it's worth the gain.

> Please also teach your editor to not re-indent code you didn't touch.
> That greatly eases review.

It didn't reindent, but it did remove tabs. I'll change that.

>> See INSTALL_windows_msys2_64.txt for tools install; there is also
>> INSTALL_windows_msys2_32.txt, which is very similar, but has "32"
>> instead of "64" in lots of places. Having two different files makes it
>> easier to cut and paste the commands.
>
> I appreciate your efforts to document the build process. However, please
> keep in mind that we don't provide half the amount of instructions on
> building on any other OS. 

I think we should; there's nothing worse than trying to follow some
install instructions only to discover there is some crucial bit of
information missing.

For example, are the Botan args to configure necessary on Debian? I just
assume so (because they are required on cygwin and msys2), but I didn't
check. If I had not done the msys2 install first, I would not have known
to use them, and would have wasted time figuring that out for Debian.

What is the downside of having explicit, complete, instructions?

On the other hand, a large part of INSTALL_windows_msys2_64.txt talks
about how to install msys2. That sort of instruction is not required for
Debian and other Linux distros, because it is widely available on the
web. It is _not_ available for msys2; I had to figure it out with help
from the mailing list.

Notice that a significant portion of the messages we've exchanged are
about "what tools are you using?". Having identical tool setups is
essential to tracking down bugs.

> Already before those additions, I felt the
> urge to merge, simplify and reduce the information into one file.

Why?

That will make it harder to follow, which means more likely to get it
wrong.

There are only 100 lines in INSTALL_windows_msys2_64.txt.

We could factor out the 'install msys2' part, but I don't see the point.

> I think exact commands should go into a script or on the wiki, if yo
> want. In the source tree, I'd say a single INSTALL_windows.txt showing
> different build options and outlining special considerations for Windows
> should suffice. 

I don't understand the rationale for moving stuff to the wiki. That's
harder to edit, and it won't be kept up to date. 

> As it stands, an average Windows user would need a guide on how to
> choose the correct INSTALL_windows_* file.

We are not addressing the average Windows user - that's my mom and
siblings, who read email but never compile anything.

We are addressing people who want to compile monotone. That's me and
you and a few others. I find these instructions necessary, and I'm doing
the work of maintaining them; what's the downside?

The guide is in INSTALL (just updated in nvm.msys2-mingw-64). I suggest
we delete INSTALL_windows_mingw.txt.

-- 
-- Stephe