injection failing using compilation server

[postite]

minimalistic example failing with compilation server. works good without

https://gist.github.com/postite/2059bc4600f8115a8469

bug or what ?

[kevinresol]

What is the error?

[postite]

hello @kevinresol .
no explicit error . but injected string will be null . ( only with compilation server ) .

[kevinresol]

Hm, maybe a minject bug?

[kevinresol]

Can you reduce it to a minimal example and see if it is minject or ufront problem

[postite]

sorry for the late reply.
my gist is already a very minimalistic ufront-setup i really don't know what to remove from here .

[postite]

btw i've just tried minject alone with compilation server and it works fine. so it's definitely a ufront bug we got here.

[postite]

fyi first two compilations work fine but then ... compilation runs ok but injected variable is null

[kevinresol]

This should be haxe-related.
Let's track it at HaxeFoundation/haxe#4892

[postite]

ok i've seen this! let the masters do their macro compiler Magic :)

[kevinresol]

@postite could you report your haxe version on that thread too?

Edit: I think Simn has figured out already. So no worry.

[kevinresol]

It turns out that it is not haxe's problem but minject's was coded in a way that is not compatible with the compilation server. I have pushed a fix to here.

That works, but it is not optimal. Because it uses onGenerate to modify the meta and onGenerate loops through all types, even they are already cached by the compilation server. Therefore I have changed onGenerate to use tink_syntaxhub (here), which will make it only process uncached types.

I will keep this open until the fix got merged into upstream minject

[jasononeil]

Thanks for looking into this @kevinresol. Let's see if the team at Massive Interactive are happy to merge. Hopefully!