bpo-40334: Support CO_FUTURE_BARRY_AS_BDFL in the new parser

[pablogsal]

Closes: we-like-parsers#113

https://bugs.python.org/issue40334

[pablogsal]

Most of the code is threading down the flags to the parser (commit
784d851) and fix a bug when inlining actions (commit 770fd11). The actual fix is implemented in commit
c970217.
In the future when we don't need to have compatibility with the old parser, we can make the tokenizer return a different token for "<>" and then intercept that in the parser in another alternative.

[gvanrossum]

Is this needed for alpha 6?

[pablogsal]

Is this needed for alpha 6?

Nop

[gvanrossum]

Thanks for all the work here for threading through the flags! I think that might solve one of my problems with type comments.

[gvanrossum]

PEP 7

Suggested change

	_PyPegen_check_barry_as_flufl(Parser *p){
	_PyPegen_check_barry_as_flufl(Parser *p) {

Also this function could use a comment explaining what it returns. Apparently nonzero is an error, meaning either <> detected in Guido mode or != detected in Barry mode, so zero means "okay", except it also could mean "exception". (I guess some later check will eventually detect the exception.)

[pablogsal]

(I guess some later check will eventually detect the exception.)

In the exception path, we can return anything IIRC because upstream code will catch that and propagate it. The return value is only used to signal if the subrule for the token should return NULL or the token (indicating that it parses).

Maybe is best to always return an error (0) even after the raise.

[gvanrossum]

Note that 0 is not an error here, it's success. (All this is confused by using strcmp(), which returns 0 for equality.) What happens if you return -1 for errors?

[pablogsal]

Maybe is best to always return an error (0) even after the raise.

Sorry, this should have been: "return an error (nonzero) even after the raise"

What happens if you return -1 for errors?

The parser reports an error because the sub-rule will return NULL for the token

[gvanrossum]

Go ahead, maybe we'll make the alpha yet!