Bicep.LangServer 0.46.1 (latest release) does not implement these three LSP
features, so iLSP now answers them itself directly in the proxy instead of
forwarding to the underlying LS:
- textDocument/signatureHelp: curated database of ~50 commonly used Bicep
built-in functions (resourceId, guid, concat, take, format, ...) with a
simple bracket-depth call-context detector (ilsp/bicep_lsp/bicep_functions.py,
ilsp/bicep_lsp/local_features.py:compute_signature_help).
- textDocument/foldingRange: stack-based bracket matcher over {}/[]/() plus
contiguous // comment-block folding (compute_folding_ranges).
- workspace/symbol: best-effort search over documents currently open in the
editor session (iLSP has no full-workspace index, only what's been opened),
matching top-level resource/module/var/param/output/func/type declarations
(compute_workspace_symbols).
The proxy now advertises these in the initialize response capabilities
(signatureHelpProvider, foldingRangeProvider, workspaceSymbolProvider) so
editors actually send the requests, and intercepts them in _client_to_ls
before they'd otherwise be silently dropped/errored by the real Bicep LS.
A send lock (_LockedConn) prevents interleaved frames now that both
_client_to_ls and _ls_to_client can write responses to the client socket.
13 new unit tests in tests/test_local_features.py. Closes the last 3 gaps
found in the full LSP capability audit.
_ls_to_client called _inject_completions() on EVERY response with an id+result,
not just textDocument/completion responses. pop_context() defaulted to
{'type': 'unknown'} for any untracked id, which _inject_completions treats as
'inject module-name completions' — so list-shaped LS responses for
definition/references/documentSymbol/formatting/codeAction were silently
replaced or prefixed with the Bicep module catalog list instead of returning
real Bicep LangServer results.
Found via a full LSP-capability audit against production: definition,
references, documentSymbol, formatting and codeAction all incorrectly
returned module-completion items instead of real results.
Fix: pop_context() now returns None when the response id wasn't recorded as
a pending completion request, and _ls_to_client only calls
_inject_completions() when a real completion context was found — every other
response now passes through untouched.
Adds support for autocomplete in multi-line array syntax like:
roles: [
'KEY_VAULT_ ← cursor triggers completion here
]
Previously only worked on same line as opening bracket:
roles: ['KEY_VAULT_ ← only this worked
Changes:
- Walk backwards up to 10 lines to find array opening (e.g. "roles: [")
- Detect if cursor is inside array based on indentation and quotes
- Stop lookback if closing bracket found (not in array anymore)
- Add test case for nested multi-line array completion
- Improve lsp_bridge.py error handling with traceback logging
- Add lsp_bridge_debug.sh wrapper for easier IntelliJ debugging
- Update EDITOR_SETUP.md with correct IntelliJ LSP4IJ config
Fixes autocomplete for deeply nested structures like:
assignments: [{ roles: ['APP_CONFIGURATION_...'] }]
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Previously the 'if lru_items:' guard meant that when our catalog returned
an empty list (e.g. a param with no allowed enum values), the Bicep LS
completions (Bicep keywords, schema types) would leak through unchanged.
Now the replace is unconditional for version/param/param_value contexts.
The Bicep LS has no knowledge of our private ACR registry so its output
in these positions is always noise — suppress it even when we have nothing
better to show.
The lookback regex used ([^']+) requiring one or more version chars,
so 'br/modules:modules/keyvault:' (no version yet) silently fell through
to the unknown context and injected all module names instead of params.
Change + to * to allow empty version string. The param_completion_items
fallback already handles empty version by picking the closest schema.
For specific LRU contexts (version, param, param_value), the Bicep LS was
appending its own random/irrelevant completions alongside the LRU catalog
items. The LS has no knowledge of the private ACR registry, so its suggestions
in these positions are noise.
Now the LS items are discarded entirely for these three contexts. LS items
are still kept (below LRU items) for module_path and unknown contexts.
