You think you can write regex?

[Jon]

I got it to compile by changing this config:

PCRE_CONFIG_LINK_SIZE

The output is an integer that contains the number of bytes used for
internal linkage in compiled regular expressions. For the 8-bit
library, the value can be 2, 3, or 4. For the 16-bit library, the value
is either 2 or 4 and is still a number of bytes. For the 32-bit
library, the value is either 2 or 4 and is still a number of bytes. The
default value of 2 is sufficient for all but the most massive patterns,
since it allows the compiled pattern to be up to 64K in size. Larger
values allow larger regular expressions to be compiled, at the expense
of slower matching.

 

But as it mentions slower matching I'll leave it as is.

[James]

Yeah, it's a crazy regex and would probably never be performed IRL. But like Mat suggested, the limitation should be documented

Out of interest Jon, how much slower is it compiling a normal regex by using the PCRE_CONFIG_LINK_SIZE setting? If it's negligible for average regular expressions, wouldn't it be good to leave on for the big'ns?

Edited August 14, 2013 by James

[Jon]

How would I document it? Do you how a 64KB compiled regex can be expressed in anything useful?

[jchd]

From ealier experiments, leave the link_size at its default of 2. Compilation and matching are much slower if this parameter is changed, due to sub-optimal internal code generation. My tests were at the time where only PCRE8 existed (UTF-8) so I don't know how it affects current library.

Anyway, the default is large enough for any reasonable pattern in real-world application IMHO.

[jchd]

Update: Pillip Hazel (PCRE author) just confirms that with linksize = 2, PCRE hits its internal limit of 64K while reaching 1589395 bytes of pattern (short 1599858 - 1589395 = 10463 pattern bytes only!) on his Linux box.

What that means is that we're fairly close to be able to handle regexps that large including an enormous number of non-capturing groups and alternations. Therefore I don't see the need to bump to linksize 3 as we would incur pointless penalty in real-world uses.

[Jon]

For the 16bit version of pcre you can only use link=2 or 4 anyway.  That number 1589395 is similar to the @error/@extended value in my orginal script - so that makes sense.

[jchd]

Yes I meant 4. Not yet used to pcre16 myself.

[Jon]

I also have regexpbuddy but it sucks by not supporting many features and clearly author completely lost interest years ago.

Spooky, it just got updated. Looks the same but works in high DPI mode and apparently lots of backend changes.

[jchd]

Yup, I saw that but I question the need to "paywarely" update.