Optimize string allocation in uri_parser_rfc3986 when getting IPv6 hosts and url paths

[LamentXU123]

Currently, php_uri_parser_rfc3986_host_read and php_uri_parser_rfc3986_path_read use the smart_str API to build the host and path strings. I think smart_str_append* introduces unnecessary overhead here, since it use repetitive boundary checks and dynamic memory reallocations.

This PR optimize string allocation in uri_parser_rfc3986 by replacing smart_str with pre-calculated zend_string_alloc. Focusing on two main functions, php_uri_parser_rfc3986_host_read() and php_uri_parser_rfc3986_path_read(), which effect getHost() and getrawHost()

for php_uri_parser_rfc3986_host_read()
This PR constructs the IPv6/IPFuture host directly using a fixed-length zend_string (formatted as [ + hostText + ] + \0). This replaces the previous smart_str appending process.

for php_uri_parser_rfc3986_path_read()
This PR first traverses the segments to pre-calculate the total length (total_len). This includes the leading / and segment delimiters. Then it performs a single zend_string_alloc() and populates the path content and delimiters using memcpy, and finally append the \0 terminator.

Benchmark script: bench.php

for getHost()

└─$ php ../../bench.php --old ~/Desktop/php/php-raw/php-src-master/sapi/cli/php --new ~/Desktop/php/opt-3/php-src-master/sapi/cli/php --iters 20000000 --mode host 
-------------------------------------------------------
|          Test | old avg(ns) | new avg(ns) | diff(%) |
-------------------------------------------------------
|    ipv6_short |         714 |         656 |   8.12% |
-------------------------------------------------------
|     ipv6_full |         806 |         746 |   7.44% |
-------------------------------------------------------
| ipv6_loopback |         743 |         671 |   9.69% |
-------------------------------------------------------
|    ipv6_mixed |         944 |         861 |   8.79% |
-------------------------------------------------------

for getrawHost()

└─$ php ../../bench.php --old ~/Desktop/php/php-raw/php-src-master/sapi/cli/php --new ~/Desktop/php/opt-3/php-src-master/sapi/cli/php --iters 20000000 --mode raw_host
-------------------------------------------------------
|          Test | old avg(ns) | new avg(ns) | diff(%) |
-------------------------------------------------------
|    ipv6_short |         557 |         509 |   8.62% |
-------------------------------------------------------
|     ipv6_full |         621 |         572 |   7.89% |
-------------------------------------------------------
| ipv6_loopback |         570 |         550 |   3.51% |
-------------------------------------------------------
|    ipv6_mixed |         667 |         679 |  -1.80% |
-------------------------------------------------------

[iluuu1994]

If this case really needs to be optimized, I would prefer extending smart_str to support appending to the string without bounds checks, with debug assertions. But I'm not code-owner.

[LamentXU123]

If this case really needs to be optimized, I would prefer extending smart_str to support appending to the string without bounds checks, with debug assertions. But I'm not code-owner.

I think this is the better solution initially. But I don't want to touch smart_str just because of this small patch.

[TimWolla]

@LamentXU123 Thank you. Can you prepare another benchmark with the latest changes? Ideally using hyperfine. The getPath() implementation would be more interesting here, since most URLs have a path, whereas IPv6 literals as hosts are more rare.

[LamentXU123]

Thank you. Can you prepare another benchmark with the latest changes?

Sure. The optimized one is 1.05x faster in this benchmark case.

<?php

$url = "http://example.com/segment1/segment2/segment3/segment4/segment5/segment6/segment7/segment8/segment9/segment10/a/b/c/d/e/f/g/h/i/j/k/l/m/n/o/p/q/r/s/t/u/v/w/x/y/z";

$sum = 0;
for ($i = 0; $i < 100000; $i++) {
    $uri = Uri\Rfc3986\Uri::parse($url);
    $path = $uri->getPath();
    $sum += strlen($path);
}

echo $sum, PHP_EOL;

└─$ hyperfine '~/Desktop/php/php-opt/php-src-master/sapi/cli/php bench.php' '~/Desktop/php/php-raw/php-src-master/sapi/cli/php bench.php'
Benchmark 1: ~/Desktop/php/php-opt/php-src-master/sapi/cli/php bench.php
  Time (mean ± σ):     592.3 ms ±  20.0 ms    [User: 498.2 ms, System: 83.7 ms]
  Range (min … max):   571.4 ms … 631.6 ms    10 runs
 
Benchmark 2: ~/Desktop/php/php-raw/php-src-master/sapi/cli/php bench.php
  Time (mean ± σ):     623.8 ms ±  16.6 ms    [User: 569.0 ms, System: 46.9 ms]
  Range (min … max):   600.4 ms … 655.4 ms    10 runs
 
Summary
  ~/Desktop/php/php-opt/php-src-master/sapi/cli/php bench.php ran
    1.05 ± 0.05 times faster than ~/Desktop/php/php-raw/php-src-master/sapi/cli/php bench.php  

In cases when the path is less nested, like:

<?php

$url = "http://example.com/segment1";

$sum = 0;
for ($i = 0; $i < 1000000; $i++) {
    $uri = Uri\Rfc3986\Uri::parse($url);
    $path = $uri->getPath();
    $sum += strlen($path);
}

echo $sum, PHP_EOL;

They are at the similar speed, the optimized one is 1.01x faster

└─$ hyperfine '~/Desktop/php/php-opt/php-src-master/sapi/cli/php bench.php' '~/Desktop/php/php-raw/php-src-master/sapi/cli/php bench.php'
Benchmark 1: ~/Desktop/php/php-opt/php-src-master/sapi/cli/php bench.php
  Time (mean ± σ):     984.1 ms ±  27.9 ms    [User: 883.8 ms, System: 83.6 ms]
  Range (min … max):   927.8 ms … 1014.5 ms    10 runs
 
Benchmark 2: ~/Desktop/php/php-raw/php-src-master/sapi/cli/php bench.php
  Time (mean ± σ):     992.0 ms ±  29.1 ms    [User: 914.3 ms, System: 62.4 ms]
  Range (min … max):   940.4 ms … 1026.5 ms    10 runs
 
Summary
  ~/Desktop/php/php-opt/php-src-master/sapi/cli/php bench.php ran
    1.01 ± 0.04 times faster than ~/Desktop/php/php-raw/php-src-master/sapi/cli/php bench.php 

[TimWolla]

I don't love the additional complexity, but it seems worth it.