They were undertested, and since #96954 might involve a
rewrite of this part of the code we want to ensure that
there won't be any behavioral change.
Co-authored-by: Carl Friedrich Bolz-Tereick <cfbolz@gmx.de>
Unicode has grown since Python first gained support for it,
when Unicode itself was still rather new.
This pair of test cases was added in commit 6a20ee7de back in 2000,
and they haven't needed to change much since then. But do change
them to look beyond the Basic Multilingual Plane (range(0x10000))
and cover all 17 planes of Unicode's final form.
This adds about 5 seconds to the test suite's runtime. Mark the
tests as CPU-using accordingly.
Having these in a separate file from the one that's named after the
module in the usual way makes it very easy to miss them when looking
for tests for these two functions.
(In fact when working recently on is_normalized, I'd been surprised to
see no tests for it here and concluded the function had evaded being
tested at all. I'd gone as far as to write up some tests myself
before I spotted this other file.)
Mostly this just means moving all the one file's code into the other,
and moving code from the module toplevel to inside the test class to
keep it tidily separate from the rest of the file's code.
There's one substantive change, which reduces by a bit the amount of
code to be moved: we drop the `x > sys.maxunicode` conditional and all
the `RangeError` logic behind it. Now if that condition ever occurs
it will cause an error at `chr(x)`, and a test failure. That's the
right result because, since PEP 393 in Python 3.3, there is no longer
such a thing as an "unsupported character".
The purpose of the `unicodedata.is_normalized` function is to answer
the question `str == unicodedata.normalized(form, str)` more
efficiently than writing just that, by using the "quick check"
optimization described in the Unicode standard in UAX #15.
However, it turns out the code doesn't implement the full algorithm
from the standard, and as a result we often miss the optimization and
end up having to compute the whole normalized string after all.
Implement the standard's algorithm. This greatly speeds up
`unicodedata.is_normalized` in many cases where our partial variant
of quick-check had been returning MAYBE and the standard algorithm
returns NO.
At a quick test on my desktop, the existing code takes about 4.4 ms/MB
(so 4.4 ns per byte) when the partial quick-check returns MAYBE and it
has to do the slow normalize-and-compare:
$ build.base/python -m timeit -s 'import unicodedata; s = "\uf900"*500000' \
-- 'unicodedata.is_normalized("NFD", s)'
50 loops, best of 5: 4.39 msec per loop
With this patch, it gets the answer instantly (58 ns) on the same 1 MB
string:
$ build.dev/python -m timeit -s 'import unicodedata; s = "\uf900"*500000' \
-- 'unicodedata.is_normalized("NFD", s)'
5000000 loops, best of 5: 58.2 nsec per loop
This restores a small optimization that the original version of this
code had for the `unicodedata.normalize` use case.
With this, that case is actually faster than in master!
$ build.base/python -m timeit -s 'import unicodedata; s = "\u0338"*500000' \
-- 'unicodedata.normalize("NFD", s)'
500 loops, best of 5: 561 usec per loop
$ build.dev/python -m timeit -s 'import unicodedata; s = "\u0338"*500000' \
-- 'unicodedata.normalize("NFD", s)'
500 loops, best of 5: 512 usec per loop
This file started life as a script, before conversion to a
`unittest` test file. Clear out some legacies of that conversion
that are a bit confusing about how it works.
Most notably, it's unlikely there's still a good reason to try
to recover from `unicodedata` failing to import -- as there was
when that logic was first added, when the module was very new.
So take that out entirely. Keep `self.db` working, though, to
avoid a noisy diff.
Hangul composition check boundaries are wrong for the second character
([0x1161, 0x1176) instead of [0x1161, 0x1176]) and third character ((0x11A7, 0x11C3)
instead of [0x11A7, 0x11C3]).
I have compared output between pre- and post-patch runs of these tests
to make sure there's nothing missing and nothing broken, on both
Windows and Linux. The only differences I found were actually tests
that were previously *not* run.
makeunicodedata.py: download all data files from unicode.org,
switch to extracting Unihan data from zip file.
Read linebreakprops and derivednormalizationprops even for
old versions, even though they are not used in delta records.
test:unicode.py: U+11000 is now assigned, use U+14000 instead.
Database (Py_UNICODE_TOLOWER, Py_UNICODE_ISDECIMAL, and others) now accept
and return characters from the full Unicode range (Py_UCS4).
The differences from Python code are few:
- unicodedata.numeric(), unicodedata.decimal() and unicodedata.digit()
now return the correct value for large code points
- repr() may consider more characters as printable.
svn+ssh://pythondev@svn.python.org/python/trunk
........
r79494 | florent.xicluna | 2010-03-30 10:24:06 +0200 (mar, 30 mar 2010) | 2 lines
#7643: Unicode codepoints VT (0x0B) and FF (0x0C) are linebreaks according to Unicode Standard Annex #14.
........
r79496 | florent.xicluna | 2010-03-30 18:29:03 +0200 (mar, 30 mar 2010) | 2 lines
Highlight the change of behavior related to r79494. Now VT and FF are linebreaks.
........
svn+ssh://pythondev@svn.python.org/python/trunk
........
r78646 | victor.stinner | 2010-03-04 13:09:33 +0100 (jeu., 04 mars 2010) | 5 lines
Issue #1054943: Fix unicodedata.normalize('NFC', text) for the Public Review
Issue #29.
PR #29 was released in february 2004!
........
CF Bolz-Tereick
Benjamin Peterson
Serhiy Storchaka
Alex Waygood
Victor Stinner
Batuhan Taskaya
Erlend Egeberg Aasland
Ammar Askar
Greg Price
Wonsup Yoon
Berker Peksag
Zachary Ware
Antoine Pitrou
Ezio Melotti
Alexander Belopolsky
Martin v. Löwis
Amaury Forgeot d'Arc
Mark Dickinson
Florent Xicluna
Victor Stinner