| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
|
|
|
|
|
|
|
|
| |
The implementation that was used for the "always last" separator
for NUMAFTER when LOCALE was turned on for non-OSX OSs was buggy.
This new implementation does not rely on string value tricks, but rather
implements a sentinel-based methodology to identify the "always last"
separator and place it last explicitly.
|
|
|
|
| |
These tests verify that the option puts numbers after non-numbers.
|
|
|
|
|
|
| |
All infrastructure has been added, but no tests have been yet been
inserted to verify that the code is working properly. That will come
in the next commit.
|
|
|
|
|
|
| |
This is in preparation for adding a few new versions of the null
string, one which will be sorted after other strings, and another
pair which will be used when locale is not enabled.
|
|
|
|
|
|
|
|
|
|
|
|
| |
Because OSX is a "dumb sorting" OS when locale is turned on, the
method by which "dumb sorting" was being tested for natcmp worked
on non-OSX OSs, but failed on OSX. This is addressed by using patch
for both True and False returns of dumb_sort.
For the test of using natcmp with an alg, if hypothesis yields very
extreme integers then conversion of those integers into a float
(due to alg.REAL) can cause the sorting to give unexpected results.
Switching to alg.SIGNED fixes this.
|
|\
| |
| | |
Add a natcmp() for comparing in Python 2.
|
| | |
|
| | |
|
| | |
|
| | |
|
| |\
| |/
|/| |
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Rather than generating testing values and then removing them in
post-processing, now test values are created correct-by-construction
using the newer .filter API of hypothesis strategies.
Whilst doing this filtering, it was found that the range_check
function was doing the unnecessary step of converting its input to
floats, so this has been removed (which simplifies testing).
Last, some of the tests for the command-line --filter arguments have
been simplified because the ROI of the thorough hypothesis tests
was low.
|
| | |
|
| | |
|
| |
| |
| |
| | |
This reverts commit fee6378d840d4fe2e36f9904ec1cee3b386d9bef.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
|/
|
|
| |
of two-tuples.
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
| |
- Added additional unicode number support for Python 3.7.
- Added information on how to install and test.
|
| |
|
| |
|
| |
|
|
|
|
|
| |
Some tests with LOCALE would fail because the given input had an
embedded null character. These are now pre-filtered.
|
| |
|
|
|
|
|
|
|
|
| |
Previously, there was no information in the README indicating
how to run the unit tests, and the documentation had incorrect
information. The instructions are now accurate.
This closes #46.
|
| |
|
|
|
|
|
| |
- Fixed ``StopIteration`` warning on Python 3.6+.
- All Unicode input is now normalized.
|
| |
|
|\
| |
| | |
Add normalization to incoming Unicode data
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The input normalization has been moved out of the "input_transform"
function (which was called in the "parse_string" function) and now
is the first step of the "parse_string" function. This is because
the data needs to be normalized even if the "input_transform" function
is skipped.
Tests have been reworked to understand this change.
|
| | |
|
|/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
All unicode input now gets 'NFD' normalization, which ensures that
all characters that look the same are represented by the same code
points. 'NFD' was chosen because it is the expanded for which will
cause (for example) 'é' to be placed immediately after 'e' rather than
after 'z'.
Users can choose 'NFKD' with ns.COMPATIBILITYNORMALIZE (or ns.CN) which
will change certain characters to their compatible (and often ASCII)
representation. This may be useful to cause force numbers in odd
representations to be transformed to ASCII which will potentially give
better sorting orders.
This will close issue #44.
|
|\ |
|
|/ |
|
|
|
|
|
|
|
| |
In order to test PyICU, we need to use the Trusty distribution because
PyICU needs -std=c++11 compiler option to gcc. To use pypy on Trusty,
sudo=True must be enable, but I do not want to use that. So, I am
removing pypy from testing.
|
|
|
|
| |
The version of pypy has been added to avoid a build failure.
|