talideon.com

Entries for December 2008

December 3, 2008 at 5:36PM Custom error handlers + the @-operator in PHP = FAIL

This is definitely up there as one of the most retarded PHP misfeatures I’ve encountered.

One of our systems in work needs to check to see if the host given in an email address can receive mail. For that, we’ve a nice short function that does exactly what RFC5821, and all the other SMTP RFCs say to do to check for this:

function can_host_receive_mail($host) {
    if (checkdnsrr($host, 'MX')) {
        return true;
    }
    if (checkdnsrr($host, 'A') || checkdnsrr($host, 'AAAA')) {
        $h = @fsockopen($host, 25, $_errno, $_errstr, 30);
        if ($h) {
            fclose($h);
            return true;
        }
    }
    return false;
}

Simple, right? Wrong. The key line here is the call to fsockopen(), which uses the @ operator to suppress the warning if the host can’t be contacted on port 25. Only the code is behaving as if the @ operator isn’t there at all.

You see, the particular piece of software where this is causing a problem has a custom error handler that intercepts PHP errors and converts them to exceptions to allow uniform error handling. However, it appears that if you’ve a custom error handler, the @ operator is rendered useless.

This is a big problem, especially with fsockopen() because not being able to connect to a host on a given port is, while a little exceptional, not an error; it’s one of these cases where you want an error code rather than an exception because coping with failure in this case is part of the normal flow of the program. That is why PHP triggers an E_WARNING rather than an E_ERROR, and why I was using the @ operator up above.

This breaks if you’re using a custom error handler, and @ stops suppressing the warning in spite of the fact that the presence of @ before an expression should logically by virtue of its locality override any PHP error (as opposed to exception) handling.

To work around this, I altered it to catch the exception the error handler throws:

function can_host_receive_mail($host) {
    if (checkdnsrr($host, 'MX')) {
        return true;
    }
    if (checkdnsrr($host, 'A') || checkdnsrr($host, 'AAAA')) {
        try {
            $h = fsockopen($host, 25, $_errno, $_errstr, 30);
        } catch (AFK_TrappedErrorException $ex) {
            return false;
        }
        if ($h) {
            fclose($h);
            return true;
        }
    }
    return false;
}

Damn, that sucks.

A custom error handler should not render the @ operator useless; it should allow suppression of the custom error handler too. The @ operator is there for a reason.

December 5, 2008 at 5:25PM PHP SimpleXML bug: prefix required for attributes in namespace other than the default namespace

I’ve just submitted PHP Bug #46769. This bit me at work because it caused the tests for the EPP client code to start failing suddenly on my machine.

The story is that the fix for Bug #43221 made prefixes on attribute qnames required if the namespace of that attribute was anything other than the default namespace.

Previously to this ‘fix’, libxml2 was let do its job and was able to figure out what the prefix was from the namespace. The ‘fix’ prevents libxml2’s ability to do this.

Here’s the example I posted up demonstrating the bug:

<?php
$ns_foo = "tag:example.com,2008:foo";
$ns_xsi = "http://www.w3.org/2001/XMLSchema-instance";

$root_doc = <<<EOT
<?xml version="1.0" encoding="UTF-8"?>
<a xmlns:xsi="$ns_xsi"
   xmlns="tag:example.com,2008"
   xsi:schemaLocation="tag:example.com,2008 root.xsd"
   xmlns:foo="$ns_foo"/>
EOT;

$root = simplexml_load_string($root_doc);
$child = $root->addChild('bar', null, $ns_foo);
$child->addAttribute('schemaLocation', "$ns_foo foo.xsd", $ns_xsi);
print_r($root->asXml());

What should be happening and what was happening previously was that because the XML schema instance namespace was declared on a parent element of <foo:bar/>, the correct document was being generated:

<?xml version="1.0" encoding="UTF-8"?>
<a xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
   xmlns="tag:example.com,2008" xmlns:foo="tag:example.com,2008:foo"
   xsi:schemaLocation="tag:example.com,2008 root.xsd">
  <foo:bar xsi:schemaLocation="tag:example.com,2008:foo foo.xsd"/>
</a>

However now this code triggers the warning “Warning: SimpleXMLElement::addAttribute(): Attribute requires prefix for namespace” and the attribute is no longer being added to <foo:bar/>.

This broke the EPP client badly because the code for building requests declares the namespaces on the root element of the document and relies on being able to reference them later by name. The upshot of this is that our EPP client (and, no doubt, other code) cannot be deployed on PHP 5.2.6 and later.

We’ll see how this is followed up. Suffice it to say that this is not the best start to the weekend, and that once again unit testing has displayed its awesomeness.

December 18, 2008 at 8:12PM Testing embedding stuff from Github Gist: pipedarg

I knocked this out earlier because I needed something to allow me to open HTML document attachments in Firefox from mutt. It writes standard input to a temporary file and then executes the remainder of the arguments passed to the script, with the path of the temporary file given as the final argument. Might be useful to somebody.

It was also a good opportunity to test out embedding a Github Gist. If you can’t see it, here, you can see the source for pipedarg here.

Update (27th December): I wasn’t thinking straight when I wrote that, and I’d corrected the script a couple of days after posting it up. You can see what changes I made from the history.

December 31, 2008 at 12:44AM Harry Potter Kata

I did the Harry Potter Kata earlier because I was feeling a little bored. Here’s the result. I use cents rather than euros as my unit of currency here.

I’d a start when I was thinking about how I’d solve the problem. I thought that being greedy and trying to make the biggest sets would do the trick, but the first example test case given soon managed to disabuse me of that notion.

Then I got the idea of turning things around 90 degrees by creating a set of buckets within the basket and distributing the sets between them. However, given the sample test case, I realised that the maximum discount could come from an obscure distribution of books, so I took the brute force route and decided to enumerate all the possible ways to distribute all the books of a certain volume, tracking which combination gave the greatest discount, and did this iteratively for each volume.

I already had the intuition to eliminate the larger groups of volumes sooner than the smaller ones, and I knew that the important thing was the number of books of each volume there were rather than what the particular volumes were, so sorting basket by volume made sense and avoided the need to do any backtracking later.

I looked at other implementations after I’d finished, and mine seems to be sorter and more general--though not purposely so--than most, but I’m still pretty sure there’s a much more elegant way to find an ideal distribution of volumes than the brute force method I used, though I’ve yet to think of one. Here’s the code:

#!/usr/bin/env python

import doctest
import itertools

def sum_basket(basket, unit_price, discount_pcs):
    """
    Gets the sum price of the books in the basket.

    basket - list of number of each book
    unit_price - price of a book
    discount_pcs - percentage discount to apply for each additional book in a set

    >>> price = 800
    >>> discount_pcs = [0, 0, 5, 10, 20, 25]
    >>> sum_basket([0, 0, 0, 0, 0], price, discount_pcs)
    0
    >>> sum_basket([0, 1, 0, 0, 0], price, discount_pcs)
    800
    >>> sum_basket([0, 3, 0, 0, 0], price, discount_pcs)
    2400
    >>> sum_basket([1, 1, 0, 0, 0], price, discount_pcs)
    1520
    >>> sum_basket([2, 1, 2, 1, 0], price, discount_pcs)
    4080
    >>> sum_basket([2, 1, 1, 1, 1], price, discount_pcs)
    3800
    >>> sum_basket([2, 2, 2, 1, 1], price, discount_pcs)
    5120
    >>> sum_basket([1, 1, 2, 2, 2], price, discount_pcs)
    5120
    >>> sum_basket([5, 5, 4, 5, 4], price, discount_pcs)
    14120
    """
    buckets = get_basket_buckets(basket)
    result = 0

    for books in sorted(basket, reverse=True):
        best = None
        for patterns in bitmap_combinations(books, len(buckets)):
            attempt = [x + y for x, y in itertools.izip(buckets, patterns)]
            attempt_sum = 0
            for bucket in attempt:
                attempt_sum += sum_bucket(bucket, unit_price, discount_pcs)
            if best is None or attempt_sum < result:
                best = attempt
                result = attempt_sum
        buckets = best

    return result

def sum_bucket(bucket, unit_price, discount_pcs):
    """
    Gets the sum cost of a set of books.

    bucket - number of unique books in a set.
    unit_price - price of a book
    discount_pcs - percentage discount to apply for each additional book in a set

    >>> price = 800
    >>> discount_pcs = [0, 0, 5, 10, 20, 25]
    >>> sum_bucket(0, price, discount_pcs)
    0
    >>> sum_bucket(1, price, discount_pcs)
    800
    >>> sum_bucket(2, price, discount_pcs)
    1520
    >>> sum_bucket(3, price, discount_pcs)
    2160
    >>> sum_bucket(4, price, discount_pcs)
    2560
    >>> sum_bucket(5, price, discount_pcs)
    3000
    """
    return bucket * unit_price * (100 - discount_pcs[bucket]) / 100

def bitmap_combinations(ones, length):
    if length == 0:
        yield []
    else:
        if ones < length:
            for tail in bitmap_combinations(ones, length - 1):
                yield itertools.chain([0], tail)
        if ones > 0:
            for tail in bitmap_combinations(ones - 1, length - 1):
                yield itertools.chain([1], tail)

def get_basket_buckets(basket):
    """
    Creates enough basket buckets to hold the maximum number of unique book sets in
    the basket.

    >>> get_basket_buckets([0])
    []
    >>> get_basket_buckets([0, 0])
    []
    >>> get_basket_buckets([1, 0])
    [0]
    >>> get_basket_buckets([0, 1])
    [0]
    >>> get_basket_buckets([0, 2])
    [0, 0]
    >>> get_basket_buckets([1, 2])
    [0, 0]
    >>> get_basket_buckets([1, 2, 3, 5, 2])
    [0, 0, 0, 0, 0]
    """
    return [0 for i in range(max(basket))]

def _test():
    doctest.testmod()

if __name__ == '__main__':
    _test()