Обсуждение: Unaccent extension python script Issue in Windows

On Mon, Mar 11, 2019 at 09:54:45PM +0530, Ramanarayana wrote:
> I went through the python script and found that the stdout encoding is set
> to utf-8 only if python version is <=2. The same needs to be done for
> python 3

If you send a patch for that, how would it look like?  Could you also
register any patch produced to the future commit fest?  It is here:
https://commitfest.postgresql.org/23/
--
Michael

Hello.

At Sun, 17 Mar 2019 20:23:05 -0400, Hugh Ranalli <hugh@whtc.ca> wrote in
<CAAhbUMNoBLu7jAbyK5MK0LXEyt03PzNQt_Apkg0z9bsAjcLV4g@mail.gmail.com>
> Hi Ram,
> Thanks for doing this; I've been overestimating my ability to get to things
> over the last couple of weeks.
> 
> I've looked at the patch and have made one minor change. I had moved all
> the imports up to the top, to keep them in one place (and I think some had
> originally been used only by the Python 2 code. You added them there, but
> didn't remove them from their original positions. So I've incorporated that
> into your patch, attached as v2. I've tested this under Python 2 and 3 on
> Linux, not Windows.

Though I'm not sure the necessity of running the script on
Windows, the problem is not specific for Windows, but general one
that haven't accidentially found on non-Windows environment.

On CentOS7:
> export LANG="ja_JP.EUCJP"
> python <..snipped..>
..
> UnicodeEncodeError: 'euc_jp' codec can't encode character '\xab' in position 0: illegal multibyte sequence

So this is not an issue with Windows but with python3.

The script generates identical files with the both versions of
python with the pach on Linux and Windows 7. Python3 on Windows
emits CRLF as a new line but it doesn't seem to harm. (I didn't
confirmed that due to extreme slowness of build from uncertain
reasons now..)

This patch contains irrelevant changes. The minimal required
change would be the attached. If you want refacotor the
UnicodeData reader or rearrange import sutff, it should be
separate patches.

It would be better use IOBase for Python3 especially for stdout
replacement but I didin't since it *is* working.

> Everything else looks correct. I apologise for not having replied to your
> question in the original bug report. I had intended to, but as I said,
> there's been an increase in the things I need to juggle at the moment.

regards.

-- 
Kyotaro Horiguchi
NTT Open Source Software Center
diff --git a/contrib/unaccent/generate_unaccent_rules.py b/contrib/unaccent/generate_unaccent_rules.py
index 58b6e7deb7..0d645567b7 100644
--- a/contrib/unaccent/generate_unaccent_rules.py
+++ b/contrib/unaccent/generate_unaccent_rules.py
@@ -45,7 +45,9 @@ if sys.version_info[0] <= 2:
     # Python 2 and 3 compatible bytes call
     def bytes(source, encoding='ascii', errors='strict'):
         return source.encode(encoding=encoding, errors=errors)
+else:
 # END: Python 2/3 compatibility - remove when Python 2 compatibility dropped
+    sys.stdout = codecs.getwriter('utf8')(sys.stdout.buffer)
 
 import re
 import argparse
@@ -233,7 +235,8 @@ def main(args):
     charactersSet = set()
 
     # read file UnicodeData.txt
-    unicodeDataFile = open(args.unicodeDataFilePath, 'r')
+    unicodeDataFile = codecs.open(
+        args.unicodeDataFilePath, mode='r', encoding='UTF-8')
 
     # read everything we need into memory
     for line in unicodeDataFile:

Hello.

At Mon, 18 Mar 2019 14:13:34 +0900 (Tokyo Standard Time), Kyotaro HORIGUCHI <horiguchi.kyotaro@lab.ntt.co.jp> wrote in
<20190318.141334.186469242.horiguchi.kyotaro@lab.ntt.co.jp>
> Hello.
> 
> At Sun, 17 Mar 2019 20:23:05 -0400, Hugh Ranalli <hugh@whtc.ca> wrote in
<CAAhbUMNoBLu7jAbyK5MK0LXEyt03PzNQt_Apkg0z9bsAjcLV4g@mail.gmail.com>
> > Hi Ram,
> > Thanks for doing this; I've been overestimating my ability to get to things
> > over the last couple of weeks.
> > 
> > I've looked at the patch and have made one minor change. I had moved all
> > the imports up to the top, to keep them in one place (and I think some had
> > originally been used only by the Python 2 code. You added them there, but
> > didn't remove them from their original positions. So I've incorporated that
> > into your patch, attached as v2. I've tested this under Python 2 and 3 on
> > Linux, not Windows.
> 
> Though I'm not sure the necessity of running the script on
> Windows, the problem is not specific for Windows, but general one
> that haven't accidentially found on non-Windows environment.
> 
> On CentOS7:
> > export LANG="ja_JP.EUCJP"
> > python <..snipped..>
> ..
> > UnicodeEncodeError: 'euc_jp' codec can't encode character '\xab' in position 0: illegal multibyte sequence
> 
> So this is not an issue with Windows but with python3.
> 
> The script generates identical files with the both versions of
> python with the pach on Linux and Windows 7. Python3 on Windows
> emits CRLF as a new line but it doesn't seem to harm. (I didn't
> confirmed that due to extreme slowness of build from uncertain
> reasons now..)

I confirmed that CRLF actually doesn't harm and unaccent works
correctly. (t_isspace() excludes them as white space).

> This patch contains irrelevant changes. The minimal required
> change would be the attached. If you want refacotor the
> UnicodeData reader or rearrange import sutff, it should be
> separate patches.
> 
> It would be better use IOBase for Python3 especially for stdout
> replacement but I didin't since it *is* working.
> 
> > Everything else looks correct. I apologise for not having replied to your
> > question in the original bug report. I had intended to, but as I said,
> > there's been an increase in the things I need to juggle at the moment.

regards.

-- 
Kyotaro Horiguchi
NTT Open Source Software Center

On Mon, Mar 18, 2019 at 09:06:09AM -0400, Hugh Ranalli wrote:
> I'm not sure I'd classify the second change as "irrelevant." Using "with"
> is the standard and recommended practice for working with files in Python.

I honestly don't know about any standard way to do anythings in
Python, but it is true that using "with" saves from a forgotten
close() call.

> At the moment the script does nothing to close the open data file, whether
> through regular processing or in the case of an exception. I would argue
> that's a bug and should be fixed. Creating a separate patch for that seems
> to be adding work for no reason.

This script runs in a short-lived context, so it is really not a big
deal to not close the opened UnicodeData.txt.  I agree that it is bad
practice though, so I think it's fine to fix the problem if there is
another patch touching the same area of the code while on it.
--
Michael

Thanks!  I have pushed this patch.  I didn't test on Windows, but I did
verify that it works with python2 and 3 on my Linux machine.

CLDR has made release 35 already, upon download of which the script
generates a few more lines in the unaccent.rules file, as attached.

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services