PHP Pitfalls

Phabricator Flavor Text (PHP)

This document discusses difficult traps and pitfalls in PHP, and how to avoid, work around, or at least understand them.

array_merge() in Incredibly Slow When Merging A List of Arrays

If you merge a list of arrays like this:

$result = array();
foreach ($list_of_lists as $one_list) {
  $result = array_merge($result, $one_list);
}

...your program now has a huge runtime because it generates a large number of intermediate arrays and copies every element it has previously seen each time you iterate.

In a libphutil environment, you can use array_mergev() instead.

var_export() Hates Baby Animals

If you try to var_export() an object that contains recursive references, your program will terminate. You have no chance to intercept or react to this or otherwise stop it from happening. Avoid var_export() unless you are certain you have only simple data. You can use print_r() or var_dump() to display complex variables safely.

isset(), empty() and Truthiness

A value is "truthy" if it evaluates to true in an if clause:

$value = something();
if ($value) {
  // Value is truthy.
}

If a value is not truthy, it is "falsey". These values are falsey in PHP:

null      // null
0         // integer
0.0       // float
"0"       // string
""        // empty string
false     // boolean
array()   // empty array

Disregarding some bizarre edge cases, all other values are truthy. Note that because "0" is falsey, this sort of thing (intended to prevent users from making empty comments) is wrong in PHP:

if ($comment_text) {
  make_comment($comment_text);
}

This is wrong because it prevents users from making the comment "0". THIS COMMENT IS TOTALLY AWESOME AND I MAKE IT ALL THE TIME SO YOU HAD BETTER NOT BREAK IT!!! A better test is probably strlen().

In addition to truth tests with if, PHP has two special truthiness operators which look like functions but aren't: empty() and isset(). These operators help deal with undeclared variables.

In PHP, there are two major cases where you get undeclared variables -- either you directly use a variable without declaring it:

function f() {
  if ($not_declared) {
    // ...
  }
}

...or you index into an array with an index which may not exist:

function f(array $mystery) {
  if ($mystery['stuff']) {
    // ...
  }
}

When you do either of these, PHP issues a warning. Avoid these warnings by using empty() and isset() to do tests that are safe to apply to undeclared variables.

empty() evaluates truthiness exactly opposite of if(). isset() returns true for everything except null. This is the truth table:

The value of these operators is that they accept undeclared variables and do not issue a warning. Specifically, if you try to do this you get a warning:

if ($not_previously_declared) {         // PHP Notice:  Undefined variable!
  // ...
}

But these are fine:

if (empty($not_previously_declared)) {  // No notice, returns true.
  // ...
}
if (isset($not_previously_declared)) {  // No notice, returns false.
  // ...
}

So, isset() really means is_declared_and_is_set_to_something_other_than_null(). empty() really means is_falsey_or_is_not_declared(). Thus:

• If a variable is known to exist, test falsiness with if (!$v), not empty(). In particular, test for empty arrays with if (!$array). There is no reason to ever use empty() on a declared variable.
• When you use isset() on an array key, like isset($array['key']), it will evaluate to "false" if the key exists but has the value null! Test for index existence with array_key_exists().

Put another way, use isset() if you want to type if ($value !== null) but are testing something that may not be declared. Use empty() if you want to type if (!$value) but you are testing something that may not be declared.

usort(), uksort(), and uasort() are Slow

This family of functions is often extremely slow for large datasets. You should avoid them if at all possible. Instead, build an array which contains surrogate keys that are naturally sortable with a function that uses native comparison (e.g., sort(), asort(), ksort(), or natcasesort()). Sort this array instead, and use it to reorder the original array.

In a libphutil environment, you can often do this easily with isort() or msort().

array_intersect() and array_diff() are Also Slow

These functions are much slower for even moderately large inputs than array_intersect_key() and array_diff_key(), because they can not make the assumption that their inputs are unique scalars as the key varieties can. Strongly prefer the key varieties.

array_uintersect() and array_udiff() are Definitely Slow Too

These functions have the problems of both the usort() family and the array_diff() family. Avoid them.

foreach() Does Not Create Scope

Variables survive outside of the scope of foreach(). More problematically, references survive outside of the scope of foreach(). This code mutates $array because the reference leaks from the first loop to the second:

$array = range(1, 3);
echo implode(',', $array); // Outputs '1,2,3'
foreach ($array as &$value) {}
echo implode(',', $array); // Outputs '1,2,3'
foreach ($array as $value) {}
echo implode(',', $array); // Outputs '1,2,2'

The easiest way to avoid this is to avoid using foreach-by-reference. If you do use it, unset the reference after the loop:

foreach ($array as &$value) {
  // ...
}
unset($value);

unserialize() is Incredibly Slow on Large Datasets

The performance of unserialize() is nonlinear in the number of zvals you unserialize, roughly O(N^2).

call_user_func() Breaks References

If you use call_use_func() to invoke a function which takes parameters by reference, the variables you pass in will have their references broken and will emerge unmodified. That is, if you have a function that takes references:

function add_one(&$v) {
  $v++;
}

...and you call it with call_user_func():

$x = 41;
call_user_func('add_one', $x);

...$x will not be modified. The solution is to use call_user_func_array() and wrap the reference in an array:

$x = 41;
call_user_func_array(
  'add_one',
  array(&$x)); // Note '&$x'!

This will work as expected.

You Can't Throw From __toString()

If you throw from __toString(), your program will terminate uselessly and you won't get the exception.

An Object Can Have Any Scalar as a Property

Object properties are not limited to legal variable names:

$property = '!@#$%^&*()';
$obj->$property = 'zebra';
echo $obj->$property;       // Outputs 'zebra'.

So, don't make assumptions about property names.

There is an (object) Cast

You can cast a dictionary into an object.

$obj = (object)array('flavor' => 'coconut');
echo $obj->flavor;      // Outputs 'coconut'.
echo get_class($obj);   // Outputs 'stdClass'.

This is occasionally useful, mostly to force an object to become a Javascript dictionary (vs a list) when passed to json_encode().

Invoking new With an Argument Vector is Really Hard

If you have some $class_name and some $argv of constructor arguments and you want to do this:

new $class_name($argv[0], $argv[1], ...);

...you'll probably invent a very interesting, very novel solution that is very wrong. In a libphutil environment, solve this problem with newv(). Elsewhere, copy newv()'s implementation.

Equality is not Transitive

This isn't terribly surprising since equality isn't transitive in a lot of languages, but the == operator is not transitive:

$a = ''; $b = 0; $c = '0a';
$a == $b; // true
$b == $c; // true
$c == $a; // false!

When either operand is an integer, the other operand is cast to an integer before comparison. Avoid this and similar pitfalls by using the === operator, which is transitive.

All 676 Letters in the Alphabet

This doesn't do what you'd expect it to do in C:

for ($c = 'a'; $c <= 'z'; $c++) {
  // ...
}

This is because the successor to z is aa, which is "less than" z. The loop will run for ~700 iterations until it reaches zz and terminates. That is, $c will take on these values:

a
b
...
y
z
aa // loop continues because 'aa' <= 'z'
ab
...
mf
mg
...
zw
zx
zy
zz // loop now terminates because 'zz' > 'z'

Instead, use this loop:

foreach (range('a', 'z') as $c) {
  // ...
}