General FAKE improvements

[matthid]

See #1083. Bah I hope its the last time now. This turned out to be way harder than imagined.

[forki]

@bjartwolf can you please verify latest alpha works for you?

[bjartwolf]

Testing now. Letting you know in a minute...

[bjartwolf]

Works fine on TeamCity.

[forki]

thx. starting proper release

[matthid]

Lets hope the best, thanks @forki for all the patience and sorry for breaking things several times (at least we now have more unit tests).

[forki]

don't worry. I'm happy that you and other people care so much about the tool.

[forki]

I updated FAKE in Paket and get this when running build on fsprojects/Paket@844dcb8

[forki]

says it cannot load a type in the assembly

[matthid]

looking into it...

[matthid]

Sadly I cannot reproduce, tried to checkout the git hash and 'paket restore -f'. Does "-pd" give the full error message?

[forki]

ok deleted .fake folder - seems to work now

[forki]

I guess we should invalidate all caches. Can you submit a PR which does that? maybe we need a cache version encoded in the file name.

[matthid]

hm, maybe something to do with included scripts, but I thought we handle them in the caching code.... Could you send me the broken assembly (out of the recycle bin)?

[matthid]

Do you mean FAKE version?

[forki]

In paket we have: https://github.com/fsprojects/Paket/blob/master/src/Paket.Core/Nuget.fs#L25

whenever we change that property paket will not take older caches

[matthid]

Ah you mean a cache version, got it. For this particular change I added a warning file so it should have invalidated already...

[matthid]

Therefore I assume something else is to blame here.

[matthid]

I think it's better to print the exception, invalidate the cache and recompile if loading the cached assembly fails, this should work without adding any version. And it works in the case that something did go wrong when writing the cache (power outrage for example). I can come up with a PR if you agree (actually I wanted to include that but I decided to feature stop after it broke again and again :D ).

[forki]

sounds good

[matthid]

By the way with this change we can finally move FAKE forwards in the sense that we can mark some ot the <AutoOpen> modules as obsolete if we want to (as people will finally see the warnings and have a opportunity to move away). Whats the status there, is this still on the wish list?