Well, one of the purpose behind dumpast was to be able to carefully
review the output of the c++ parser on some 'real' life sources.
I managed to find some (unplanned) parsing issue due to the fuzzy
nature of the parser (it doesn't know about types):
---
1)
CTag* newtag = new CTag(searchresult->taglist[i]->tag);
Is not parsed as a declaration, but an expression were CTag and newtag are
variable (a*b)
2)
CBarShader CPartFile::s_LoadBar(PROGRESS_HEIGHT);
Is parsed as a declaration of a function pointer instead of a declaration
with a constructor (assumes that PROGRESS_HEIGHT is a type).
---
The first issue is particularly annoying. None of thoses issues have
simple fixes (would require to make the parser knowing about type in some
way). While the parser is great for prototyping, I would trust production
work on it.
That being said, it still usefull for prototyping and did provide
interesting input on how to mix preprocessing, c++ parsing and refactoring.
I started looking into ANTLR to implement a serious parser with type
handling last week. I'm currently toying with the tinyc grammar to get
features similar to our framework for ast manipulations and ast to source
conversion.
Baptiste.
|
