[RFC] Deprecation of uniqid()

2018-05-11 12:36 GMT+02:00 Alice Wonder alice@librelamp.com:

2018-05-10 16:33 GMT+02:00 Niklas Keller me@kelunik.com:

Hey,

I hereby propose to deprecate uniqid(). There have been attempts to
fix
it
(
https://wiki.php.net/rfc/uniqid), but those were rejected during
discussion, because there's no possible fix without breaking BC.
Instead
of
a subtle BC break, this RFC favors the deprecation and moving
users to
other functions.

It's to be discussed whether the function should be removed with
PHP 8.0
or
just deprecated to avoid fully breaking things where it's not
strictly
necessary. A deprecation will probably avoid most new usages,
which is
the
main goal.

RFC: https://wiki.php.net/rfc/deprecate-uniqid

Kind Regards,
Niklas

--

Hello,

as a userland user of this function I do disagree with it's outright
removal. It has it's uses.
What can be done with it is drop the $more_entropy flag and make it
generate at least as long strings and use random_bytes under the
hood for
a
better random.
It can also adopt a length parameter so you can vary the random
part as
much as you need it.

You don't always need a truly random token - I have a system that uses
uniqid to generate tens of thousands tokens per request and it's
actually
a
good thing they are time based at the start of it with a random
part at
the
end (as I said the random part should be improved and get rid of that
stupid dot when generating with $more_entropy = true).

It seems to me that for your use case, you could just use the time()
function to get part of your unique id and then use libsodium to
generated
a nonce for the "random" part, using sodium's function for increment
the
nonce between each use.

Predictable, sure, but your use case says they don't need to be a truly
random token - just unique (essentially a non-random nonce) but with
a time
component.

--

Hello Alice,

Sure, there is lots I can do about that project, including what you have
described. One thing though - client does not need it or want it or
want's
to pay for that work. That whole project is a poster child for a "side
project on a bare minimum, but done by a competent developer instead
of a
student so it actually works in the long run"

Tell the client they can use this for free.

function compat_uniqid(string $prefix='', bool $more_entropy = false)
{
static $nonce = null;
if(is_null($nonce)) {
$nonce = random_bytes(SODIUM_CRYPTO_SECRETBOX_NONCEBYTES);
}
$m = microtime(true);
$return = sprintf("%8x%05x",floor($m),($m-floor($m))*1000000);
if($more_entropy) {
sodium_increment($nonce);
$x = hexdec(substr(bin2hex($nonce),0,8));
$x = str_pad($x, 12, "0", STR_PAD_LEFT);
$return = $return . substr($x, 0, 1) . '.' . substr($x, -8);
}
return $prefix . $return;
}

slightly better if block

if($more_entropy) {
sodium_increment($nonce);
$x = hexdec(substr(bin2hex($nonce),0,12));
$return = $return . substr($x, 2, 1) . '.' . substr($x, -8);
}

Obvious patterns in the "more entropy" but the output in only suppose to
be unique, not random.

My point is that since it is trivial to create a function that is
compatible with the php function for backwards compatibility, the
problematic function in php itself doesn't need to exist.

For an actual unique id, code should be updated to generate an actual
nonce (either with pRNG every call or first time and increment if
predictable is okay) rather than use a function that clearly fails to
always produce something unique. So the function should go.

If I had a vote I would vote for the RFC and deprecate it.

2018-05-11 12:36 GMT+02:00 Alice Wonder alice@librelamp.com:

2018-05-10 16:33 GMT+02:00 Niklas Keller me@kelunik.com:

Hey,

I hereby propose to deprecate uniqid(). There have been attempts to
fix
it
(
https://wiki.php.net/rfc/uniqid), but those were rejected during
discussion, because there's no possible fix without breaking BC.
Instead
of
a subtle BC break, this RFC favors the deprecation and moving
users to
other functions.

It's to be discussed whether the function should be removed with
PHP 8.0
or
just deprecated to avoid fully breaking things where it's not
strictly
necessary. A deprecation will probably avoid most new usages,
which is
the
main goal.

RFC: https://wiki.php.net/rfc/deprecate-uniqid

Kind Regards,
Niklas

--

Hello,

as a userland user of this function I do disagree with it's outright
removal. It has it's uses.
What can be done with it is drop the $more_entropy flag and make it
generate at least as long strings and use random_bytes under the
hood for
a
better random.
It can also adopt a length parameter so you can vary the random
part as
much as you need it.

You don't always need a truly random token - I have a system that uses
uniqid to generate tens of thousands tokens per request and it's
actually
a
good thing they are time based at the start of it with a random
part at
the
end (as I said the random part should be improved and get rid of that
stupid dot when generating with $more_entropy = true).

It seems to me that for your use case, you could just use the time()
function to get part of your unique id and then use libsodium to
generated
a nonce for the "random" part, using sodium's function for increment
the
nonce between each use.

Predictable, sure, but your use case says they don't need to be a truly
random token - just unique (essentially a non-random nonce) but with
a time
component.

--

Hello Alice,

Sure, there is lots I can do about that project, including what you have
described. One thing though - client does not need it or want it or
want's
to pay for that work. That whole project is a poster child for a "side
project on a bare minimum, but done by a competent developer instead
of a
student so it actually works in the long run"

Tell the client they can use this for free.

function compat_uniqid(string $prefix='', bool $more_entropy = false)
{
static $nonce = null;
if(is_null($nonce)) {
$nonce = random_bytes(SODIUM_CRYPTO_SECRETBOX_NONCEBYTES);
}
$m = microtime(true);
$return = sprintf("%8x%05x",floor($m),($m-floor($m))*1000000);
if($more_entropy) {
sodium_increment($nonce);
$x = hexdec(substr(bin2hex($nonce),0,8));
$x = str_pad($x, 12, "0", STR_PAD_LEFT);
$return = $return . substr($x, 0, 1) . '.' . substr($x, -8);
}
return $prefix . $return;
}

slightly better if block

if($more_entropy) {
sodium_increment($nonce);
$x = hexdec(substr(bin2hex($nonce),0,12));
$return = $return . substr($x, 2, 1) . '.' . substr($x, -8);
}

Obvious patterns in the "more entropy" but the output in only suppose to
be unique, not random.

If you don't need the output to be exact same format, this avoids
collisions and is faster.

function cryptoUniqid(string $prefix = '', bool $prng = false): string
{
static $nonce = null;
if($prng || is_null($nonce)) {
$nonce = random_bytes(16);
} else {
sodium_increment($nonce);
}
$m = microtime(true);
$return = sprintf("%8x%05x", floor($m), ($m-floor($m))*1000000);
$return = $return . '.' . base64_encode($nonce);
return $prefix . $return;
}

slightly better if block

if($more_entropy) {
sodium_increment($nonce);
$x = hexdec(substr(bin2hex($nonce),0,12));
$return = $return . substr($x, 2, 1) . '.' . substr($x, -8);
}

Obvious patterns in the "more entropy" but the output in only suppose to
be unique, not random.

The most important issue in current uniqid() is it could have duplicate
even with $more_entropy=true, since it has

<time_part> + <random based on time>

structure.
System time could be adjusted backwards by ntp, etc.

However changing output by default as follows is enough for it.

<time_part> + <random based on CSPRNG>

uniqid() is one of the most abused function in PHP.
I don't have strong opinion against uniqid() deprecation, though.

Regards,

--
Yasuo Ohgaki