This is follow up to D93393.
Without this patch clangd takes the symbol definition from the static index if this definition was removed from the dynamic index.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Unit Tests
Event Timeline
I also want to propose the similar patch for fuzzyFind() (as a part of this patch or a separate one), but I faced the following problem:
Test.cpp
#define FOO 1
This example creates empty dynamic index for main file symbols and static index for preamble symbols (FOO). The location of FOO is inside Test.cpp, so (according to the logic that the static index could be stale) we should toss FOO from the static index (i.e. preamble symbols), but this behaviour is incorrect... Any advice?
clang-tools-extra/clangd/index/FileIndex.cpp | ||
---|---|---|
434 | We do not need this change to fix behaviour for removed definition, but this is the only one place where we use URI instead of path as a key. |
Ugh, this case manages to fall through the cracks of our model every time.
Originally the idea was:
- features traverse the AST for the current file, and consult the index for headers
- uh... macros do something analogous to that
- this lines up with the preamble/parsedAST split
But macros defined in the preamble section aren't in headers but also aren't parsed in parsedAST, so they're really awkward to handle. We end up playing whack-a-mole with these bugs. (See loadMainFilePreambleMacros in CodeComplete.cpp).
In this case, it seems reasonable to include the macros in the dynamic index for the file. This is a bit of work (requires handling MainFileMacros::Name in SymbolCollector::handleMacros I guess) but is probably less work than excluding the symbols from the index entirely and having to work around this in every feature.
It *also* seems reasonable to exclude these main-file symbols from the preamble index. For one thing, the preamble index will include all the files with any symbol defined in indexedFiles(), and the whole file isn't indexed!
I'm not 100% sure this bug would block adding the equivalent change for fuzzyFind though - it'll affect documentSymbol but not code-completion IIUC (because of the workaround I mentioned above). And these symbols are pretty rare.
clang-tools-extra/clangd/index/FileIndex.cpp | ||
---|---|---|
434 | Yeah... nevertheless the key is totally arbitrary here, so we might as well use whatever is most convenient/cheapest. | |
clang-tools-extra/clangd/index/Merge.cpp | ||
81 | this seems OK because we expect the definition to see the canonical declaration - subtle enough to be worth a comment. |
With the similar change for fuzzyFind(), WorkspaceSymbols.Macros test fails. But maybe I can create one more review for this and we will discuss it there.
clang-tools-extra/clangd/index/FileIndex.cpp | ||
---|---|---|
434 | It's not arbitrary after D93393 (we use these keys to create the list of indexed files). But as it turns, the problem is deeper... I will try to explain. void test(); test.c #include "test.h" void te^st {}
With this change only 1 element will be returned (definition of test in test.c and nothing from test.h) I am not sure what we can do here, but I think we need some special merge for preamble and main file indexes (another merge class for this case instead of MergedIndex?) |
Sorry, incomplete thought. I meant workspace/symbols would be broken for such macros, but maybe that was acceptable (it's not a terrible common feature nor symbol kind). Not great but we're trading one bug for another.
In particular, if we plan to fix both I don't think the *sequencing* matters.
clang-tools-extra/clangd/index/FileIndex.cpp | ||
---|---|---|
434 |
Hmm, what if we made the return value of the indexedFiles functor a bitmask instead of a single boolean (contains symbols, contains refs, ...)? |
Ok. I will propose a separate patch for fuzzyFind() similar to this one.
Thank you for review.
clang-tools-extra/clangd/index/FileIndex.cpp | ||
---|---|---|
434 | This sounds reasonable. I will try to implement a solution for this problem based on your suggestion (in a separate patch) |
We do not need this change to fix behaviour for removed definition, but this is the only one place where we use URI instead of path as a key.