Rootmodule diagnostics

[appilon]

This refactor based on the new decoder architecture brings the language server as close as possible to having project wide diagnostics #270 . Diagnostics are calculated for the entire rootmodule on file open and change. It does mean that a user that has several rootmodules open won't see diagnostics far a particular rootmodule (effectively folder) until a file is opened within a that rootmodule.

Closes #270

[appilon]

Previously I had this set to HCL, however now I think since the diagnostics are coming from the new decoder, they could be interpreted as not just syntax, but also provider schema specific? Therefore I thought saying the source was Terraform was most appropriate, let me know if that is an incorrect interpretation

[radeksimko]

The spec doesn't seem to go into much detail in describing this field, but it seems to suggest that each linter should be reported with its own Source - in our case I reckon that would apply to tflint and I would think potentially to terraform validate too?

The key question is how does the field affect the UX or anything in practice - I have no idea 🤷‍♂️

Re HCL/Terraform - our current diags in particular actually come just from hclsyntax.ParseConfig, so HCL might be more appropriate here. The schema is not involved in the validation at all.

terraform-ls/internal/terraform/rootmodule/root_module.go

Lines 464 to 468 in 8bfea75

	f, pDiags := hclsyntax.ParseConfig(src, name, hcl.InitialPos)
	diags = append(diags, pDiags...)
	if f != nil {
	files[name] = f
	}

[appilon]

I'll switch it back then.

[radeksimko]

LGTM, just one question about the subject-less diags.

[appilon]

@radeksimko I moved uri to it's own pkg, I stopped suppressing subjectless diagnostics, and simply ensured a zero'd range is provided (So in theory a diagnostic will be attributed to the file, but will default to first line/column, diagnostics must have a range in LSP). PTAL

[radeksimko]

Was this added as a reaction to any particular edge case you ran into?

[appilon]

HCL range start with index 1, in the event there is no subject, I passed a zero'd HCL range to the converter, this would result in negative values in the LSP conversion (since LSP range is index 0), this just ensures we don't send an invalid range

[ghost]

I'm going to lock this issue because it has been closed for 30 days ⏳. This helps our maintainers find and focus on the active issues.
If you have found a problem that seems similar to this, please open a new issue and complete the issue template so we can capture all the context necessary to investigate further.