repo2rlenv-swe-next / tasks /swe-next-6f59fc5cd8372ff0ed91 /environment /source /tests /test_formatutils.py
Download tasks/swe-next-6f59fc5cd8372ff0ed91/environment/source/tests/test_formatutils.py from FineEnvs/repo2rlenv-swe-next: direct link, hf CLI and curl.
- Browser
- Download file 2.06 kB
-
https://huggingface.co/datasets/FineEnvs/repo2rlenv-swe-next/resolve/main/tasks/swe-next-6f59fc5cd8372ff0ed91/environment/source/tests/test_formatutils.py
- Command line
-
hf download hf://datasets/FineEnvs/repo2rlenv-swe-next/tasks/swe-next-6f59fc5cd8372ff0ed91/environment/source/tests/test_formatutils.py
-
curl -L -o test_formatutils.py https://huggingface.co/datasets/FineEnvs/repo2rlenv-swe-next/resolve/main/tasks/swe-next-6f59fc5cd8372ff0ed91/environment/source/tests/test_formatutils.py
2.06 kB
| import re | |
| from collections import namedtuple | |
| from boltons.formatutils import (get_format_args, | |
| split_format_str, | |
| tokenize_format_str, | |
| infer_positional_format_args, | |
| DeferredValue as DV) | |
| PFAT = namedtuple("PositionalFormatArgTest", "fstr arg_vals res") | |
| _PFATS = [PFAT('{} {} {}', ('hi', 'hello', 'bye'), "hi hello bye"), | |
| PFAT('{:d} {}', (1, 2), "1 2"), | |
| PFAT('{!s} {!r}', ('str', 'repr'), "str 'repr'"), | |
| PFAT('{[hi]}, {.__name__!r}', ({'hi': 'hi'}, re), "hi, 're'"), | |
| PFAT('{{joek}} ({} {})', ('so', 'funny'), "{joek} (so funny)")] | |
| def test_pos_infer(): | |
| for i, (tmpl, args, res) in enumerate(_PFATS): | |
| converted = infer_positional_format_args(tmpl) | |
| assert converted.format(*args) == res | |
| _TEST_TMPLS = ["example 1: {hello}", | |
| "example 2: {hello:*10}", | |
| "example 3: {hello:*{width}}", | |
| "example 4: {hello!r:{fchar}{width}}, {width}, yes", | |
| "example 5: {0}, {1:d}, {2:f}, {1}", | |
| "example 6: {}, {}, {}, {1}"] | |
| def test_get_fstr_args(): | |
| results = [] | |
| for t in _TEST_TMPLS: | |
| inferred_t = infer_positional_format_args(t) | |
| res = get_format_args(inferred_t) | |
| assert res | |
| def test_split_fstr(): | |
| results = [] | |
| for t in _TEST_TMPLS: | |
| res = split_format_str(t) | |
| results.append(res) | |
| assert res | |
| def test_tokenize_format_str(): | |
| results = [] | |
| for t in _TEST_TMPLS: | |
| res = tokenize_format_str(t) | |
| results.append(res) | |
| assert res | |
| def test_deferredvalue(): | |
| def myfunc(): | |
| myfunc.called += 1 | |
| return 123 | |
| myfunc.called = 0 | |
| dv = DV(myfunc) | |
| assert str(dv) == '123' | |
| assert myfunc.called == 1 | |
| assert str(dv) == '123' | |
| assert myfunc.called == 1 | |
| dv.cache_value = False | |
| assert str(dv) == '123' | |
| assert myfunc.called == 2 | |
| assert str(dv) == '123' | |
| assert myfunc.called == 3 | |