[PATCH] D158413: [Lex] Introduce Preprocessor::LexTokensUntilEOF()

[Vassil Vassilev via Phabricator via cfe-commits]

v.g.vassilev added a comment.

This looks reasonable to me, but I'd wait for @aaron.ballman greenlight as he seems to have had more comments/ideas.



================
Comment at: clang/lib/Lex/Preprocessor.cpp:1000
+  std::vector<Token> toks;
+  while (1) {
+    Token tok;
----------------
Hahnfeld wrote:
> aaron.ballman wrote:
> > v.g.vassilev wrote:
> > > aaron.ballman wrote:
> > > > Hahnfeld wrote:
> > > > > aaron.ballman wrote:
> > > > > > I'd prefer not to assume the token stream has an EOF token (perhaps the stream is one only being used to parse until the `eod` token instead), so if we can turn this into a non-infinite loop, that would make me more comfortable.
> > > > > I'm not sure I understand entirely. Do you want something like
> > > > > ```
> > > > > tok.isOneOf(tok::unknown, tok::eof, tok::eod)
> > > > > ```
> > > > > instead of `tok.is(tok::eof)`? Can this happen at the level of the `Preprocessor`?
> > > > I was thinking something more along the lines of:
> > > > ```
> > > > if (Tokens) {
> > > >   for (Token Tok; !Tok.isOneOf(tok::eof, tok::eod); Lex(Tok))
> > > >     Tokens->push_back(Tok);
> > > > }
> > > > ```
> > > > but I hadn't thought about `tok::unknown`; that might be a good one to also include given that clangd operates on partial sources.
> > > > 
> > > I was wondering if we could somehow merge this routine with `Parser::SkipUntil` since they seem to be doing a very similar tasks.
> > That could perhaps end up looking reasonable (they do similar tasks aside from collecting the tokens that are being skipped). Do you need the interface to be on `Preprocessor` or `Parser` though (or does it not really matter for you)?
> > `tok.isOneOf(tok::unknown, tok::eof, tok::eod)`
> 
> I implemented this check, let me know if this looks reasonable. The code you posted doesn't do what we need because we also want to lex if `Tokens` is `nullptr`, so the hierarchy must be an `if` inside the loop.
> Given these additional token kinds, does `UntilEOF` still make sense or do we want another name? Note that I'll leave `repl_input_end` to https://reviews.llvm.org/D158415.
> 
> > I was wondering if we could somehow merge this routine with `Parser::SkipUntil` since they seem to be doing a very similar tasks.
> 
> I'm not sure this makes sense, given that `Parser::SkipUntil` requires some knowledge about the structural input. At the very least, I'd prefer not to go into that direction for this change.
I am not sure I understand the reasoning but I somewhat see that having Parser's `SkipUntil` be implemented with our new `Preprocessor::LexUntil...` would require a lot more work. How about adding a fixme note capturing this as a future possible refactoring?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D158413/new/

https://reviews.llvm.org/D158413