build-and-deploy(i686): address the dreaded `.dll` base address problem

[dscho]

I just tried to deploy the i686 version of the MSYS variant of curl, but ran into problems.

This PR hopefully fixes those problems.

[dscho]

Let's see whether this run (with the proposed fixes in effect) succeeds...

[dscho]

Let's see whether this run (with the proposed fixes in effect) succeeds...

Didn't. I got the wrong path for the Git for Windows SDK. Here is the next attempt.

[dscho]

Here is the next attempt.

That failed, too, with the same Cwd.dll problem. I kicked off a re-run in the hopes that it will "magically" be resolved. If it is not, I will have to use action-tmate and play with the base address to find one that works. This is so tedious, I really look forward to dropping i686 support.

[dscho]

I kicked off a re-run in the hopes that it will "magically" be resolved.

Nope.

[dscho]

I will have to use action-tmate and play with the base address to find one that works.

Nope.. Apparently the Gods are not with me today.

[dscho]

I finally think that I know what is going on. Something lets pacman.exe think that it should run rebase.exe after installing the packages. And that rebase messes things up that would otherwise have worked. Will keep digging.

[dscho]

We finally seem to have successfully jumped the autoreconf hurdle.

[dscho]

The build worked, at long last, and the i686 flavor of the curl package is successfully deployed.

So I force-pushed away the debug commits and will mark this as ready for review.

[rimrul]

Hopefully we won't need to do this anymore after the issues with Cygwin 3.4 get worked out.

[dscho]

Hopefully we won't need to do this anymore after the issues with Cygwin 3.4 get worked out.

Well, Cygwin v3.4.x does not even support i686 anymore... So git-sdk-32 will be stuck with Cygwin v3.3.x forever ;-)

[rimrul]

Right. I've somehow missed that. So the whole dynamicbase support doesn't help us at all. Damn.