Repository navigation
Conversation
* Allow console or diagnostics reporting * Make SourceLocation track ranges * Introduce ScopeTree * Implement ScopeTree finding for variables * Implement DiagnosticReporter * Introduce Compiler::analyze * Use strings in Report instead of chars * Remove newline endings from reporter calls * Introduce SourceFileLoader * Introduce SemanticTokensBuilder * Add -fPIC to antlr * Introduce Report[er]::clear * Allow workspace module building
* SemanticTokensBuilder updates - A single lexer token can be multiple semantic tokens (eg. multi- line string) * Add ScopeTree to grab local variables * Change FunctionCall::parameters to return a ptr This internal error was previously never triggered, but now since we no longer "fast fail", it can occur. * Null CompilerWorkspace checks A CompilerWorkspace cannot be created if the source cannot be found. * Make module load failure an error, not a fatal This change is needed to continue diagnostics after failing to locate a dependent module. * Use parse tree visitor for semantic tokens * Fix nullptr issues with failed parsing * Fix memory issue with SourceFileIdentifier
* Make include load failure an error, not a fatal * Updates to ScopeTree * Update find_constant * Address review comments - Remove newline
* Begin work to support completion * Address review comments
* Add continue_on_error for analysis * Fix compiler warning on assignment in conditional * Fix formatting
… condition (#606) * add uninit, false, true keywords for sem tokens * add missing continue_on_error check
… SourceFile tokens (#617) * Add token_index to Range * Use SourceFile token_stream to bypass lexer reset
…e strings (#621) * Replace CRLF with LF * No need to replace CRLF since its handled in the loop * Remove duplicate code
…l path for module path names (#624) * Keep track of replaced Node from Optimizer * Use `Clib::FullPath` on module path names
… within extension (#627) * Add support for formatting ranges * Formatting * Fix NO_PCH compilation * Really fix NO_PCH compilation... * Fix colmn offsets
* Add nullptr checks in AST generation * Fix adding new line at EOF with ranged formatting * Changes needed for hover and completion support * Refactor for proper scope tree query * Additional nullptr checks * More scope tree query enhancements * Add support for class member and method scope queries * Address Discord discussions - Make `find_*` an exact match
Fixes an issue with ShortCircuitWarning visiting a FunctionCall whose children are nullptrs due to the FunctionCall having taken its arguments via `fc.take_arguments()`.
…alysis mode (#988) * Keep a missing module or include fatal outside analysis The diagnostics work turned the report.fatal() for an unreadable module (use_module) and include (locate_include_file) into report.error() plus return, in every mode. ecompile then kept going and printed follow-on errors about the missing functions and constants. BuilderWorkspace now carries analysis_mode. Only Compiler::analyze() sets it; compile_file_steps and ExpressionEvaluator pass false. The two failures are fatal again unless analysis_mode is set. vscode-escript keeps the error-and-continue behaviour it needs to analyse the rest of the file. * Keep the first declaration of a duplicated class method class_declaration() replaced the class's method FunctionLink each time a method name repeated. master skips a name that is already in the methods map (#808), but the rebase of the diagnostics work dropped that check. ecompile is unaffected, because the duplicate definition is an error that stops compilation before the analyzer runs. Under Compiler::analyze() the link pointed at the last declaration instead of the first. * Print the source line under a short-circuit warning again The diagnostics work dropped the source line that ecompile prints below "Short circuit change: ... will potentially not been executed". vscode-escript does not want the line, because the editor already marks the location. ShortCircuitReporter now takes analysis_mode. Outside analysis it appends the line as master does. Under Compiler::analyze() the message stays a single line. * Keep the final newline when a format-off block runs to EOF The newline that FormatterInsertNewlineAtEOF adds moved into PrettifyLineBuilder::mergeEOFNonTokens(). It is skipped when _rangeSkippedNewlineAtEOF is set, and mergeRawContent() set that flag for any skipped block that reached the last line. So without a format range, a file ending in an unclosed "// format-off" lost its final newline under ecompile -F and -Fi. setSkipLines() now also receives whether a format range is active, and only then does reaching EOF set the flag. vscode-escript's format-range requests behave as before. * Record a class method's location even without a parameter list class_declaration() added the method to all_function_locations only when functionParameters() existed. master records it for every method. ecompile never sees the difference, because a missing parameter list is a parse error and AST building stops after it. Under analysis with continue_on_error, a method whose parameter list fails to parse was left out, so the analyzer's "See also" hints could not point at it. * Build debugger expressions without parse-error recovery ExpressionEvaluator, which evaluates expressions for the DAP debugger, built its workspaces with continue_on_error set. master's evaluator does no such recovery. A parse error already throws "Invalid expression" before the AST is built. So the flag only changed an expression that parses but hits an unhandled AST case: it evaluated to an uninitialized value instead of failing with an error. Both workspaces now pass false. * Name the module in build_module()'s load errors build_module() passed its "Refusing to load" and "Unable to load" messages to report.error() as separate pieces ("Unable to load '", pathname, "'."). report.error() takes a format string, and fmt drops the unused arguments, so the message was cut off before the path. Both now use "{}". * Skip a case group without labels in the Disambiguator visit_case_dispatch_group called selectors.children.back(). The grammar (switchLabel+) and the error checks between compile stages keep that list non-empty in ecompile. Compiler::analyze() runs disambiguate() after parse errors, and recovery on a half-typed `case ( elem ) 5: x` builds a group with no label, so the language server crashed. * Build ScopeTree and unoptimized nodes only in analysis mode ecompile filled CompilerWorkspace::scope_tree on every compile and the optimizer kept every folded Identifier as Node::unoptimized_node. Only vscode-escript reads either, through Compiler::analyze(). scope_tree is now a unique_ptr created only in analysis mode. LocalVariableScope and SemanticAnalyzer::analyze() skip it when null. The optimizer keeps unoptimized_node only when analysis_mode is set. Outside analysis mode, globals.remove_all_but( 0 ) is no longer called, as on master. * Mark the declarations only vscode-escript uses Each declaration that only Compiler::analyze() or the formatter's range mode reaches carries "Analysis only: used by vscode-escript, not by ecompile or pol.", so a single grep lists them all. The comment on SourceFileLoader::get_contents says why it is virtual. * Give the analysis helpers internal linkage Move starts_with, with_base_classes and define() into anonymous namespaces. Convert to unsigned char before std::tolower, which is undefined for negative char values. Remove the unused SemanticTokenType2 and reformat six branch files. * Modernize the compiler's analysis helpers Use range-for with structured bindings, starts_with, contains and std::ranges algorithms. Replace std::list and unique_ptr-wrapped returns with std::vector and std::optional. Have SemanticToken::append_from_lexer_token append to the token vector instead of returning a heap-allocated list. Take strings by const reference and copy no parameter vector in visit_function_call.
Collaborator
|
coverage: 81.209% (-0.8%) from 81.995% — ecompile-diagnostics-140b0dc4 into master |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Kevin is to blame ;)
Brings the compiler work behind the vscode-escript language
server (diagnostics, semantic highlighting, completion,
signature help, formatting) to master. This is the
long-running ecompile-diagnostics branch, rebased onto
current master. Code used only for analysis runs only under
Compiler::analyze(), so ecompile and the server compile
scripts as they do on master.
The branch combines these PRs, each reviewed against the
diagnostics branch:
Continue-on-Error condition
getting SourceFile tokens
multi-line strings
canonical path for module path names
range) within extension
branch outside analysis mode
Five later commits have no PR of their own: regular
expression literals, FunctionCall children gathering, uninit
class functions, enum class / scoped constants, and
short-circuit evaluation warnings.
Companion: polserver/vscode-escript#132