php.net |  support |  documentation |  report a bug |  advanced search |  search howto |  statistics |  random bug |  login
Bug #46408 Locale number format settings can cause pg_query_params to break with numerics
Submitted: 2008-10-28 19:40 UTC Modified: 2013-01-14 20:26 UTC
Votes:30
Avg. Score:4.7 ± 0.6
Reproduced:27 of 28 (96.4%)
Same Version:26 (96.3%)
Same OS:24 (88.9%)
From: alec at smecher dot bc dot ca Assigned: lstrojny (profile)
Status: Closed Package: PostgreSQL related
PHP Version: 5.*, 6 OS: *
Private report: No CVE-ID: None
View Developer Edit
Welcome! If you don't have a Git account, you can't do anything here.
If you reported this bug, you can edit this bug over here.
(description)
Block user comment
Status: Assign to:
Package:
Bug Type:
Summary:
From: alec at smecher dot bc dot ca
New email:
PHP Version: OS:

 

 [2008-10-28 19:40 UTC] alec at smecher dot bc dot ca
Description:
------------
Using PHP's setlocale function can cause number formats to change (e.g. formatting floats using a comma instead of a decimal, such as 3,5 instead of 3.5). This causes pg_query_params to break when binding floats into the parameter array, as psql doesn't understand comma-formatted numbers.

You may need to generate the hr_HR locale on your system in order for number formatting to work as described above once setlocale has been called (see the debug "echo" in the reproduce code).

I reproduced the problem with 5.2.6 and 5.2.7RC2.

Reproduce code:
---------------
<?php

ini_set('display_errors', E_ALL);
setlocale(LC_ALL, 'hr_HR.utf-8', 'hr_HR');

$number = 3.5;
$dbuser = 'putdbusernamehere';
$dbpass = 'putdbpasswordhere';

echo "Three and a half is: " . $number . " (should come out as 3,5)\n";

$conn = pg_connect("host=localhost user=$dbuser password=$dbpass");
$result = pg_query_params($conn, 'SELECT $1::numeric', array($number));

pg_close($conn);

?>


Expected result:
----------------
Three and a half is: 3,5 (should come out as 3,5)

Actual result:
--------------
Three and a half is: 3,5 (should come out as 3,5)

Warning: pg_query_params(): Query failed: ERROR:  invalid input syntax for type numeric: "3,5" in /home/asmecher/cvs/ojs2-stable/test.php on line 13


Patches

Pull Requests

History

AllCommentsChangesGit/SVN commitsRelated reports
 [2008-10-28 22:54 UTC] alec at smecher dot bc dot ca
In case it wasn't clear, the bug IMO is that pg_query_params is using a means to convert doubles to strings that is potentially incompatible with the database.
 [2008-10-31 18:28 UTC] alec at smecher dot bc dot ca
FYI, there's a discussion of the same bug, which also appeared (in a separate implementation) in the implementation of the Pear::DB package: <http://pear.php.net/bugs/bug.php?id=3021>
 [2008-11-18 22:59 UTC] lsmith@php.net
<RhodiumToad> lsmith: in a parameterized query it's always wrong to use 
locale-specific delimiters

RhodiumToad is also known as Andrew Gierth and is a highly respected 
expert on #postgresql on freenode.

As such I will reopen the bug ..
 [2008-11-18 23:16 UTC] alec at smecher dot bc dot ca
Thanks, lsmith and RhodiumToad. FYI, this bug also exists in PDO (I can post reproduce code if it's helpful).
 [2008-11-21 13:09 UTC] jani@php.net
I guess it's an issue always if extension does 'convert_to_string()'.
Easily avoided in code: Only do setlocale() prior to outputting stuff.
And then restore the locale right after output. :)
 [2009-07-26 18:59 UTC] jerico dot dev at gmail dot com
@jani: When I pass in a double, I expect pg_query_params() to prepare it in a way that can be understood by the database independent of my locale settings. AFAIK the implementation of pg_query_params() is also inconsistent with that of the mysql driver which correctly accepts double typed parameters independent of locale.

I guess you were not entirely serious when you proposed that one should switch the locale before using pg_query_params(), were you?
 [2010-05-14 12:13 UTC] lewis at peppermind dot de
This issue also appears with pg_execute(), when passing float values to the bind array, with a locale that uses comma as decimal separator (such as de_DE and most other european locales).
 [2012-03-31 05:43 UTC] yohgaki@php.net
-Assigned To: +Assigned To: yohgaki
 [2012-03-31 05:43 UTC] yohgaki@php.net
I guess setting locale to database locale resolve this issue.
 [2012-04-17 08:21 UTC] yohgaki@php.net
-Status: Assigned +Status: Wont fix
 [2012-04-17 08:21 UTC] yohgaki@php.net
It seems it does not work that way.

http://www.postgresql.org/docs/8.4/static/locale.html
Check that PostgreSQL is actually using the locale that you think it is. The 
LC_COLLATE and LC_CTYPE settings are determined when a database is created, and 
cannot be changed except by creating a new database. Other locale settings 
including LC_MESSAGES and LC_MONETARY are initially determined by the 
environment the server is started in, but can be changed on-the-fly. You can 
check the active locale settings using the SHOW command.


You may get currency as your locale setting, but not numbers.

yohgaki@[local] ~=# select 123456789.12345::text::money;
          money           
--------------------------
 EUR12.345.678.912.345,00

yohgaki@[local] ~=# select 123456789.12345::text::numeric;
     numeric     
-----------------
 123456789.12345

PHP's pgsql is getting all data as string. If there is a method that returns 
numbers as you expected, then you'll get your outputs needed. Can do that in 
PHP.
 [2012-04-17 16:22 UTC] alec at smecher dot bc dot ca
yohgaki, I'm not sure I'm following, but to be clear: the bug entry is not about whether the output is 3,5 or 3.5. The bug is that PostgreSQL throws a syntax error when passing a floating-point number into a parameterized query. See RhodiumToad's quoted comment.

You're quoting the following psql query:
select 123456789.12345::text::numeric;

What I'm trying to point out is this:
select 123456789,12345::text::numeric;
 [2012-04-17 19:33 UTC] yohgaki@php.net
I've tried different locale see if it works and it doesn't.

I also found this
http://www.postgresql.org/docs/8.4/static/locale.html
So , cannot replaced by .

There is other case transparent usage is not allowed. Money is one of them.
http://www.postgresql.org/docs/8.4/static/datatype-money.html

If postgres is going to support conversion, then it is going to work, since it 
requires server side settings to work. (i.e. Database initialization, table 
definition) It is not a bug or feature that should be fixed in PHP.

It's possible to do locale conversion in module, it would not be a good idea in 
general.
 [2012-04-17 19:52 UTC] alec at smecher dot bc dot ca
PostgreSQL is hard-coded to parse numeric values using "." as the decimal point (see https://github.com/postgres/postgres/blob/master/src/backend/utils/adt/numeric.c in set_var_from_str for implementation).

This agrees with the SQL92 grammar for a numeric literal:
http://savage.net.au/SQL/sql-92.bnf.html#exact%20numeric%20literal

...so by my reading PHP needs fixing.
 [2012-04-17 20:06 UTC] yohgaki@php.net
>PostgreSQL is hard-coded to parse numeric values using "." as the decimal point 
(see 
>https://github.com/postgres/postgres/blob/master/src/backend/utils/adt/numeric.
>c in set_var_from_str for implementation).

The number *may* be a string, what we should do then?

Is it "3.5" or "3,5"? Only the PHP programmer who is writing the code knew which 
should be. Therefore, this won't fix.
 [2012-04-17 20:15 UTC] alec at smecher dot bc dot ca
That's not correct. In the supplied example, the data is provided to pg_query_params as numeric data. The number to string conversion is done within pg_query_params (here, I think: https://github.com/php/php-src/blob/master/ext/pgsql/pgsql.c#L1739).
 [2012-04-17 21:36 UTC] yohgaki@php.net
PostgreSQL users are supposed to pass/recieve data via strings *always* with 
C/C++ API.

Programmers are responsible how it's passed.
As as mentioned already, system will not know how numbers should be formatted 
for certain column. The code even don't care the column types of tables.

Think,

create table test (a text);

and 

insert into test (a) values ("number");

The number format is decided by PHP programmer, not pgsql module.
 [2012-04-18 01:30 UTC] alec at smecher dot bc dot ca
Your example of storing a string-formatted double in a varchar column strikes me as an unusual case involving a type mismatch, and can be worked around clearly and logically by passing "$number" in rather than $number.

If this isn't going to be fixed, there should at least be documentation in the manual for pg_query_params: "If binding numeric data for use as numeric data, make sure you first cast it to a string using '.' for a decimal separator." But that makes no sense at all to me.

Meanwhile, this same issue has been solved elsewhere as I've suggested.

Gnome: https://bugzilla.gnome.org/show_bug.cgi?id=650016
Pear::DB: http://pear.php.net/bugs/bug.php?id=3021

It's also implemented in these other cases with specific number formatting code as I believe it should:
Pear::MDB2: http://svn.php.net/viewvc/pear/packages/MDB2/trunk/MDB2/Driver/Datatype/Common.php?view=markup#l1498
mysqli: https://github.com/php/php-src/blob/master/ext/mysqli/mysqli_api.c#L102

"PostgreSQL users are supposed to pass/recieve data via strings *always* with 
C/C++ API." -- this is also incorrect, if I understand you properly. See the paramTypes column documented here http://www.postgresql.org/docs/9.1/static/libpq-exec.html for the PQexecParams function that you're using.
 [2012-04-18 02:19 UTC] yohgaki@php.net
IIRC, MDB2 cares data types. (which I think it's a design problem for loosely 
type langs.) Therefore, it may change behavior according to locale.

Anyway, It's not a matter of argument. It's the way it works. As I explained 
repeatedly, ALL params are passed as string and types are NEVER cared in C API. 
To know the data type, module should issue an additional query to get meta data. 
Additional query for each query is severe performance hit.

Therefore, it is a PHP programmer's responsibility (or Perhaps, PostgreSQL. If 
they would like to change behavior via client environment vars. I guess they 
would not.)

If you could submit patch that works for all data types and does not require 
additional query, I'll review it.
 [2012-04-18 02:26 UTC] yohgaki@php.net
BTW, you are reading PostgreSQL manual wrong.

libpq's functions never care about data types, but the server is.

If you are curious still, try to make patch that meets the requirement I've wrote.
 [2012-04-18 02:58 UTC] alec at smecher dot bc dot ca
I believe pg_query_params is broken until this is resolved, but it looks like we're not going to agree on it. I hope someone else can speak up if they do think this is a bug.

Since we disagree on the approach any patch I write to correct it will be rejected.

I'll add a comment to the manual page for pg_query_params to document this.
 [2012-04-18 03:13 UTC] yohgaki@php.net
You misunderstand how libpq/PostgreSQL works.

If you think you can make proper patch for this, clone git source and send pull 
request.
No one will stop you from that.
 [2012-04-18 12:44 UTC] claude dot pache at gmail dot com
@yohgaki (and others)

I think that, the root of the problem is the way PHP uses
the locale information, which I consider deeply broken.
Here are the  details:

In my understanding, the locale information is useful only
for *output*, i.e. for messages destined to the user.
They should not be used for any internal conversion
from one type to the other, unless the result is
destined to output.

The problem is, that PHP uses the locale for any
automatic conversion from number to string.
This behaviour is ok in the following case:

    echo "Three and a half is: " . $number;

However, in the following cases, this is NOT correct,
because the resulting string must not be localised:

    * constructing a JSON object (I hope that json_encode()
      does NOT use internal number-to-string conversion);
    * using bcmath package (I have personnaly be bitten by
      this misfeature);
    * construct a SQL request (the present case);
    * etc.

In all these cases, you have to do one of the following options:
    (1) never use any locale other than en_US
       (and re-implement manually the locale feature);
    (2) carefully check the type of each and every parameter
        and explicitely perform a correct conversion when needed,
        e.g. using number_format(..., '.', '');
    (3) fix PHP to NOT use locale for number-to-string conversion
        unless it is explicitely asked for
       (side note: historically, there has been a similar
       problem with the "magic quote" misfeature);
    (4) modify the modules bcmath, postgresql, etc,
       so that they circumvent the mentionned PHP misfeature,
       i.e., they do the  option (2) above at your place.

In my dreams, the option (3) would be implemented,
but pragmatically, I think that option (4)
has more chance to be implemented rapidly, if ever.

I think that alec asked precisely the option (4) to be implemented.
(Personnally, I have opted for option (1).)

Claude

P.S. The option (4) might seem a non-optimal hack.
However, do not forget that programming languages
and API should be adapted to the needs of the programmers,
and not the other way round.
 [2012-09-06 18:03 UTC] alec at smecher dot bc dot ca
I suggest this patch.

diff -u -r php-5.4.6/ext/pgsql/pgsql.c php-5.4.6-mod/ext/pgsql/pgsql.c
--- php-5.4.6/ext/pgsql/pgsql.c	2012-08-14 21:26:05.000000000 -0700
+++ php-5.4.6-mod/ext/pgsql/pgsql.c	2012-09-06 10:59:45.000000000 -0700
@@ -23,6 +23,7 @@
 /* $Id$ */
 
 #include <stdlib.h>
+#include <locale.h>
 
 #define PHP_PGSQL_PRIVATE 1
 
@@ -1736,7 +1737,15 @@
 			} else {
 				zval tmp_val = **tmp;
 				zval_copy_ctor(&tmp_val);
+
+				// PSQL requires . for radix; convert to string,
+				// avoiding problems with doubles and locales
+				// using , as a radix character instead
+				// (see https://bugs.php.net/bug.php?id=46408)
+				char *current_locale = setlocale(LC_NUMERIC, "C");
 				convert_to_string(&tmp_val);
+				setlocale(LC_NUMERIC, current_locale);
+
 				if (Z_TYPE(tmp_val) != IS_STRING) {
 					php_error_docref(NULL TSRMLS_CC, E_WARNING,"Error converting parameter");
 					zval_dtor(&tmp_val);
 [2012-09-06 18:28 UTC] alec at smecher dot bc dot ca
Pull request filed at https://github.com/php/php-src/pull/186
 [2013-01-14 20:26 UTC] lstrojny@php.net
-Status: Wont fix +Status: Closed -Assigned To: yohgaki +Assigned To: lstrojny
 [2013-01-14 20:26 UTC] lstrojny@php.net
The fix for this bug has been committed.

Snapshots of the sources are packaged every three hours; this change
will be in the next snapshot. You can grab the snapshot at
http://snaps.php.net/.

 For Windows:

http://windows.php.net/snapshots/
 
Thank you for the report, and for helping us make PHP better.

Fixed in 5.5 and master.
 
PHP Copyright © 2001-2024 The PHP Group
All rights reserved.
Last updated: Thu Nov 21 11:01:29 2024 UTC