haxe
haxe copied to clipboard
HaxeFoundation
[display] partial check for modified files
We have disabled checks for files modifications in display mode, which is good for hx files since they are sent by VSHaxe. However this cause some problems for registerModuleDependency, when the target file is processed by some macro which build a model.
Exemples : CastleDB, Heaps resources, Heaps Texts Builder, etc.
In that case, when making for instance structural changes to CDB, the changes won't reflect in completion until a new manual build was triggered, which is not a very good user experience compared to previous behavior.
Could we then reenable filetime checks (and corresponding module rebuild) for files that are either registerModuleDependency (or / and) not .hx files?
We could also consider sending file change events for everything on the IDE side, not just .hx files, if Haxe can do something with that.
Not sure what's best here. I think only considering .hx files is a quick solution. However, there could be an external dependency to a .hx file as well which would then still have this problem. A better solution would be special treatment for the registerModuleDependency modules, but that requires some internal cleanup because we're currently using MFake for other things (at least import.hx files).
We could also consider sending file change events for everything on the IDE side, not just .hx files, if Haxe can do something with that.
That should work in principle, but I'm worried about getting spammed with server/invalidate requests, e.g. when .dump files are created or with the C++ target.
Test disabled because of Win32 failure: https://dev.azure.com/HaxeFoundation/GitHubPublic/_build/results?buildId=950&view=logs&j=e61806a3-e319-5a7c-85d5-a2d0ec789e63&t=962e5de9-78fb-5a57-b35d-753d148002b2&l=446
I have disabled this test again because it has a high failure rate on our CI. I don't know where the randomness comes from here and why it's so bad for this test in particular.