On Friday, 7 October 2022 at 00:57:38 UTC, rassoc wrote:
On 10/7/22 01:39, torhu via Digitalmars-d-learn wrote:
regex is about ten times faster then.
Interesting! Using your code, I'm seeing a 1.5x max difference
for ldc, nothing close to 10x. Welp, the woes of superficial
benchmarking. :)
Benchmark results depend on many things, such as the actual text
in both needle and haystack and the needle length. Are we dealing
with unicode text by the way? One example is searching for
something like "äußere" in
https://www.gutenberg.org/ebooks/6343.txt.utf-8
If it's the source code, then searching for
"sqlite3_value_bytes16" in the sqlite3.c file from
https://www.sqlite.org/2022/sqlite-amalgamation-3390400.zip may
be a good test too.
I'm getting at least 5x difference in favor of regex with LDC on
these two examples.
Also are we allowed to artificially construct needle and haystack
to blow up this test rather than only benchmarking it on typical
real data?