Bug in jLocale, when an accent is in utf-8 format.

Hello,

fixed it, works fine.

Because I did a complete rewrite of jBundle::_loadResources(), I need to make more tests (didn't test with latin1 strings for example)

I also added a new feature, that makes possible to have multi-line rendered strings like this (end with 2 \) :

mystring = This a string that will be rendered\\
on two lines, even in HTML if you use nl2br jtpl plugin

This is cool, because you can have a big bunch of text in only one localized string, and don't have to create multiple strings just to have rendered line breaks.

Of course, it's still possible to write a string on multiple lines like before :

mystring = This a string that will be rendered\
on one line, even if it's declared on 2 lines in the property file

I also kept the spaces at the end of line, because I think it could be useful sometimes.

Last, I think that performance is improved, but didn't make any test.

This patch is not final, but I you have comments, they're welcome.

beta patch, need some more test, but if you have comments....

I don't think that loading the properties file in an array is better than reading line after line, in term of performance and memory consumption.

I have no real idea about performance gain or loss, I think I can do tests for that.

But with file_get_contents, the file is not kept open while we parse it. I think it could be nice.

I made some tests :

launched 1000 times theses scripts in cli :

<?php

$time = microtime();

$strings = array();

foreach(explode("\n",file_get_contents('errors.UTF-8.properties')) as $linenumber=>$line){
    $strings[$linenumber] = $line;
}

echo memory_get_usage(),"\t",microtime()-$time,"\n";
?>

<?php

$time = microtime();

$strings = array();

$fp = fopen('errors.UTF-8.properties','r');

$linenumber = 0;

while (!feof($fp)) {
    if($line=fgets($fp)){
        $strings[$linenumber] = $line;
        $linenumber++;
    }
}

fclose($fp);

echo memory_get_usage(),"\t",microtime()-$time,"\n";
?>

<?php

$time = microtime();

$strings = array();

foreach(file('errors.UTF-8.properties') as $linenumber=>$line){
    $strings[$linenumber] = $line;
}

echo memory_get_usage(),"\t",microtime()-$time,"\n";
?>

the average results are :

75688 0,00026674 file_get_contents()

71112 0,00034062 fgets()

75364 0,00025454 file()

so file_get_contents uses more memory but is faster (ok, we're talking about microseconds :)). Maybe file() is better.

fopen(), fgets(), fclose() use less memory, and need a little more time but nothing problematic.

So OK, if the fact that the file is kept open while parsing it is not a problem (I think it's ok because we open it read-only in fact), I'll switch back to fgets()

Ok, let's use file function.

good patch ;)

Here's the final patch.

It's working fine.

As I said before, it's now possible to end a line with \\ to get an \n in the string. So I'll update the documentation when the patch is landed in 1.0.4.

• Careful, many extra whitespace at end of some lines. Remove them.
• Your new reader don't allow same syntax as before.
  • a "#" in a value begin a comment, except when a \ is before. We cannot do that in your version.
  • we cannot specify only a whitespace in the value, by using "\w"
  • html entities shouldn't be converted. The value shouldn't be modified. So no html_entity_decode.
• When I tried to run unit tests : "Fatal error: Allowed memory size of 134217728 bytes exhausted (tried to allocate 13 bytes) in lib/jelix/core/jSelector.class.php on line 483"

more tests to check syntax in properties file

I added some few tests in unit-tests which check more syntaxic features. Your patch should pass all this test without modification. Include this patch in your patch.

Hello,

ok, I need to take a deeper look at complete syntax rules. Is there some kind of specs about the syntax of properties file somewhere ? (yes I could find these by deep analysing all the regexp in the current version, but ... ;) )

I think your new unit tests will also help in that way, thanks.

More to come in the next days.

In fact, I just realize that this ticket is invalid. jLocale works perfectly :-) In fact, this is the documentation which is uncomplete.

In fact, if we want to use a # in a property, we have to add a \ before it.

So, in the example of the ticket should be :

  titre_page = Titre momentan&\#233; 

I'm going to update the documentation about all the syntax.

I commit also unit tests attached to this ticket, with a minor correction in this tests.

Ok I didn't know about that spec ;)

I'll open a new enhancement ticket, because I think that it's quite useful to be able to have line breaks in strings : #569