fixed performance regression in endsWith()

[firewave]

This regression was introduced in 8abab5c.

std::equal() is supposed to be faster than std::string::compare() and a benchmark even supports this: https://quick-bench.com/q/JkxxUMqC8wLufOw7AUI-l1K-n7Q. But somehow in our code it leads to a big performance regression when using the Clang compiler.

before

Benchmark 1: ./simplecpp -q -D__GNUC__ -D__x86_64__ -DHAVE_CONFIG_H -I ../../xrdp/build ../../xrdp/common/file.c
  Time (mean ± σ):     166.8 ms ±  29.3 ms    [User: 142.4 ms, System: 22.6 ms]
  Range (min … max):   131.9 ms … 218.9 ms    12 runs

after

Benchmark 1: ./simplecpp -q -D__GNUC__ -D__x86_64__ -DHAVE_CONFIG_H -I ../../xrdp/build ../../xrdp/common/file.c
  Time (mean ± σ):      11.5 ms ±   0.6 ms    [User: 9.0 ms, System: 2.6 ms]
  Range (min … max):    10.9 ms …  14.6 ms    216 runs

With GCC this also exists but is much less severe.

[firewave]

Strangely I can no longer reproduce the issue.

[firewave]

Strangely I can no longer reproduce the issue.

The newly added callgrind step in the CI also shows that there is no regression with the change in question:

g++ - 75,902,589 -> 76,116,419
clang++ - 71,211,551 -> 71,311,304