The list of temporary compile command json files generated on build events can have many duplicates. This PR deduplicates that list before building up the compilation db.
This can lead to massive reductions in size (100x in my case). The raw list was something like 500k files, with only 3600 of them being unique. The resulting compile_commands.json is substantially smaller -- with duplicates, I've seen it as large as 11GB for my configuration; with this change, it is 166MB
clangd doesn't cope well with a large and heavily duplicated json file, and tends to run out of memory.
The list of temporary compile command json files generated on build events can have many duplicates. This PR deduplicates that list before building up the compilation db.
This can lead to massive reductions in size (100x in my case). The raw list was something like 500k files, with only 3600 of them being unique. The resulting
compile_commands.jsonis substantially smaller -- with duplicates, I've seen it as large as 11GB for my configuration; with this change, it is 166MBclangddoesn't cope well with a large and heavily duplicated json file, and tends to run out of memory.