https://issues.dlang.org/show_bug.cgi?id=4483

Dmitry Olshansky <[email protected]> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[email protected]

--- Comment #18 from Dmitry Olshansky <[email protected]> ---
(In reply to Walter Bright from comment #11)
> > You fail to recognize that it's broken from the begging.
> 
> and I know that std.regex suffered serious
> slowdowns because of it.

This turned out to be factually wrong, after I've spent a year and a half
constructing a non-decoding version of std.regex for no significant gain ;) A
brief exercise with a profiler shows decoding to be ~0.5% in a recent version,
as Martin points out a single 100% predictable comparison is not a problem. 

The only case where not decoding is faster is bulk-mode operations that can
take advantage of SIMD or auto-vectorization, such as:
a) Skipping comments (i.e. looping until '*' is hit then check for '/')
b) Comparing strings or searching for substring.

--

Reply via email to