Repository navigation
Non-existent files are resolved as the working directory file #573
Description
AST nodes that aren't located in any file are reported in the database as being located inside the working directory file.
As a result, the directory that the parser was called from appears in the file manager in the web UI.
This happens because cc::parser::FileLocUtil::getFilePath(const clang::SourceLocation&) returns an empty string if no file containing the SourceLocation can be found. This "path" is directly passed to cc::parser::SourceManager::getFile(const sdt::string&) in numerous places in the code. This function resolves it against the working directory, returning the same directory.
My planned solution is to make SourceManager::getFile return nullptr if a relative path (which an empty string is) is passed as the argument. This means the location_file field for these nodes will be null in the database. Does that sound like a good format?
I wanted to discuss this before implementing a fix as it requires a (small) change to the database schema.
Activity
mcserep commentedon Oct 17, 2022
@andocz Can you gives us a MWE to test and validate this issue on? (An AST node without location?)
Where did you find this unintended behaviour?
andocz commentedon Oct 19, 2022
TinyXML2 is one example. I parsed it by invoking
CodeCompass_parserwhile cd'd to CodeCompass'sbindirectory, and these nodes had thebindirectory as their location:But any C++ code should do the trick. The key is to invoke the parser from a directory unrelated to the project being parsed.
mcserep commentedon Oct 19, 2022
@andocz Thanks for the quick response, we will reproduce this first.
However I don't think that the correct behaviour would be to add
nullvalue for thelocation_filefield for these records, only if the relatedastValuereally belongs to no file. On the screenshot you shared this is not the case for many occurrences, e.g. themain(int, const char **)::TestUtil::~TestUtil()should belong to a file, I guess to the one containing themainfunction.So the source if this faulty behaviour should be examined first.
ervin7mk commentedon Dec 17, 2022
I've succesfully reproduced the issue brought up by @andocz . I created a directory unrelated to the project named ParseFromHere and I invoked the parser while inside this directory. As seen in the attached screenshots, there are nodes related to this directory.
However, there are also CodeCompass related files in the "File" table whose situation I am not sure about. Should they also appear or not?
ervin7mk commentedon Jan 16, 2023
Hi. As we previously discussed on our CodeCompass meeting, I followed the approach @andocz presented and I found that this didn't solve the issue. This code snippet is really short but you can find it at the following link: ervin7mk@203f59b
Is there an error I'm commiting or maybe the problem is solvable at a larger scale?
mcserep commentedon Feb 12, 2023
Hi @ervin7mk,
It worked for me, are you sure you made a clean parse (
-fflag), so all files are reparsed?However, a segmentation fault occurred during parsing, so most likely the returned
nullptrwas dereferenced somewhere. Usages of theSourceManager::getFile()method should be checked.ervin7mk commentedon Feb 13, 2023
Hi @mcserep ,
Oh, I wasn't aware of the
-fflag. Now I get the segmentation fault message as well. This explains a lot.Thanks! I will proceed with checking all uses of
SourceManager::getFile().ervin7mk commentedon Feb 16, 2023
Hi All!
I've checked all occurences of
SourceManager::getFile()and it looks like the folder (ParseFromHere) from which I parsed the project no longer appears in the file manager in CodeCompass, neither in the database. Please review my code and let me know if anything more needs to be added.Link to commit: 9e4f9b9
Screenshots: