Repository navigation
Replies: 2 comments 3 replies
|
For SystemVerilog, I would set expectations in two layers. The indexer does have a SystemVerilog grammar/spec. The current spec extracts things like: So the database should contain useful structural information from the files. What it does not appear to have is a full HDL elaboration/dataflow engine. SystemVerilog is also not in the current full Hybrid LSP list, so cross-file type-aware resolution is not at the same level as Python, TypeScript, Go, C/C++, Java, Kotlin, Rust, etc. That means "trace this signal from point A to point B" can fail even when parsing worked, especially if the path depends on module elaboration, generate blocks, interface/modport wiring, macros, parameterized instances, or nontrivial hierarchy. One thing worth checking: the current built-in extension table maps So my read is: If you can share one small |
|
Thank you for the SystemVerilog experiments, and sorry this did not receive a maintainer reply sooner. The comments identify a concrete need for preprocessing, a top module, and configuration-aware signal relationships; I cannot confirm complete signal tracing or a chosen Slang integration from this discussion. If you have a small shareable example now, please include the defines/top module and the expected signal path. That is enough to frame a focused request without asking you to build a full integration first. |
Uh oh!
There was an error while loading. Please reload this page.
Hi I am using this tool for systemverilog but would like to know how much information is added to the database. Tree-sitter parses alot of information from the code, what is being pulled out into the database. I have been playing with it and it's good but sometimes it says it can't trace signals from point it point? Thus the question.
All reactions