[mypyc] Add a str.format specializer which only supports empty brackets

[97littleleaf11]

Description

This PR adds a str.format specializer which only supports empty brackets. Also, the detection code can deal with the bracket literals, for example, "{{}}".

The specializer first split the formatting string by "{}". Then it replaces the brackets by PyObject_Str. Finally, all these separated substrings would be sent to a C helper function to generate a joint one.

Test Plan

• An IRbuild test
• A run test

mypyc_benchmark:

running str_format_format_method
..........
interpreted: 0.597254s (avg of 5 iterations; stdev 1.7%)
compiled:    0.312174s (avg of 5 iterations; stdev 2.9%)

compiled is 1.913x faster

[JukkaL]

I'm glad to see the performance improvement! Left some mostly minor comments. Once you handle {{ and }} this seems enough features for a first iteration.

[97littleleaf11]

Add several optimization:

• Skip empty literals
• Skip str_op on str_rprimitive

running str_format_format_method
..........
interpreted: 0.576623s (avg of 5 iterations; stdev 0.47%)
compiled: 0.280255s (avg of 5 iterations; stdev 0.63%)

compiled is 2.057x faster

[JukkaL]

The build is failing. Can you have a look?

[97littleleaf11]

@JukkaL I fixed it by replacing "{{" with "{"

[JukkaL]

Thanks for the updates! Left various suggestions about more test coverage. Right now test coverage is still a bit limited, and there may be some bugs still lurking in the code.

[JukkaL]

Please add more tests. We want fairly complete test coverage to make sure nothing has regressed. It's important both valid and error cases produce the same output when optimized. For example, any exceptions should be identical.

Examples (but there are probably more):

• Simplest possible case: ''.format()
• Plain text only: 'xyz'.format()
• Extra positional arguments: 'a{}b'.format(1, 2, 3)
• Too few arguments
• Various format specifiers that are supported within {...} (these should still work, even if they aren't optimized)
• Keyword arguments to format
• *args to format
• Invalid format string

[97littleleaf11]

Extra positional arguments case this compile error: error: Not all arguments converted during string formatting

[JukkaL]

These test cases could still be useful. Can you double check if they aren't covered:

• '{}{}'.format(1, 2) (empty literals)
• 'x{}y{}'.format(*args) (*args)

[JukkaL]

Please add unit tests for this, since it has pretty tricky logic. Also consider various error cases and edge cases.

[JukkaL]

Thanks for the updates! This is almost ready to merge. Just a few remaining comments.

[JukkaL]

Add a few more IR build test cases, such as these:

1. an empty literal
2. {{ and }} in the format string

[JukkaL]

These test cases could still be useful. Can you double check if they aren't covered:

• '{}{}'.format(1, 2) (empty literals)
• 'x{}y{}'.format(*args) (*args)

[JukkaL]

LGTM!