iconv bug - incorrect conversion

[ssj4maiko]

Description

The following code:

<?php
$string = 'Relámpago';
echo iconv('UTF-8', 'ASCII//TRANSLIT', $string)."\n";

Resulted in this output:

Rel'ampago

But I expected this output instead:

Relampago

---

Running iconv via CLI, the value is properly converted.

iconv -f 'UTF-8' -t 'ASCII//TRANSLIT' iconvinput.php

<?php
$string = 'Relampago';
echo iconv('UTF-8', 'ASCII//TRANSLIT', $string)."\n";

$ iconv -V
iconv (Ubuntu GLIBC 2.39-0ubuntu8.3) 2.39

PHP Version

PHP 8.3.4

Operating System

Ubuntu 24.04 (Docker container webdevops/php-apache-dev)

[youkidearitai]

https://www.php.net/manual/en/function.iconv.php#refsect1-function.iconv-parameters

From reference, seems depends on the system's iconv() implementation.

If and how //TRANSLIT works exactly depends on the system's iconv() implementation (cf. ICONV_IMPL). Some implementations are known to ignore //TRANSLIT, so the conversion is likely to fail for characters which are illegal for the to_encoding.

It's successful when constants ICONV_IMPL is libiconv, glibc is failed.

[cmb69]

Indeed, I assume this is not a bug in PHP, but rather depends on the iconv library and maybe its version.

@ssj4maiko, please post the output of

<?php
var_dump(ICONV_IMPL, ICONV_VERSION);

[ssj4maiko]

string(8) "libiconv"
string(4) "1.17"

So it's an external library, considering it existed on Linux, I imagined it was something kinda like a system call or something

I think I should point out, I tried many other accented characters, and all of them worked alright, this is the only one that led to such behavior, so it's not a case where the method is not working at all.

Interactive shell

php > $string = 'Relámpago'; echo iconv('UTF-8', 'ASCII//TRANSLIT', $string)."\n";
Rel'ampago
php > var_dump(ICONV_IMPL, ICONV_VERSION);
string(8) "libiconv"
string(4) "1.17"

Edit:
Strange...

php > $string = 'ÁbçdêfghÍJk LmñO'; echo iconv('UTF-8', 'ASCII//TRANSLIT', $string);
'Abcd^efgh'IJk Lm~nO

It previously worked for me, how the heck?

[ssj4maiko]

ok, here are new results that I just noticed:

folder (main)$ php iconvinput.php 
Relampago AbcdefghIJk LmnO
string(5) "glibc"
string(4) "2.39"

folder (main)$ docker compose exec -w /app http_server php iconvinput.php 
Rel'ampago 'Abcd^efgh'IJk Lm~nO
string(8) "libiconv"
string(4) "1.17"

so my host is using glibc and it works, libiconv is failing however... Although I was pretty sure I was testing everything from inside the container, I guess I wasn't.

My host PHP version is also 8.3.6, unlike the container's 8.3.4

[cmb69]

I get the same results on Windows using libiconv 1.15. I assume these transliterations are deliberate, but in any way not related to PHP. Feel free to file a bug report upstream.