[regex] detection logic based on Mobile_Detect.json

[serbanghita]

Hi,

I'd suggest improving the detection getting the detection logic from https://github.com/serbanghita/Mobile-Detect/blob/master/Mobile_Detect.json

It's just an idea, we made that .json to allow people to better port the script. We update it also on every release.

[lkarsten]

Hi. Thanks for the suggestion. Great idea to pull the ruleset into a separate file. This is a fairly low-effort kind of script, so I don't think we'll spend the effort importing your ruleset.

If anyone wants to write a VCL translator for your data, I would be happy to link to it though.

[serbanghita]

@lkarsten cool, we had quite a few requests about this, I will find someone to link to this project. Thanks for your response!

[willemk]

@serbanghita I made a little php translate script to do this: https://github.com/willemk/varnish-mobiletranslate

I've tested it with ~ 20 UA strings and it works good so far.

[serbanghita]

@willemk - Holy crap that was fast! Nice! :beers:

Btw, at Mobile_Detect we have a lot of UnitTests that keep the library relevant on each update (see https://github.com/serbanghita/Mobile-Detect/blob/master/tests/UA_List.inc.php ). Maybe you can use the tests exported to JSON at https://github.com/serbanghita/Mobile-Detect/blob/master/tests/ualist.json - it's in an unformatted format.

@willemk - can you make a README for noobs for the project https://github.com/willemk/varnish-mobiletranslate - I want to list it on our README also. Thanks!!!

[lkarsten]

@willemk Thanks, I've added a link to your generator in the README.rst file now.

Since we're generating the VCL anyway, I would love if someone took the time to optimise the output regexes a bit.

Doing massive case-insensitive regex checks for every request has do add some extra delay. For example check the most common ones first, or do a two-level matching by building an initial check on only the first characters or something similar.

I wrote a time-to-first-byte measurement tool at some point (https://github.com/lkarsten/httpttfb ) that perhaps can be used to put numbers on this.

[serbanghita]

@lkarsten are you refering to the https://github.com/willemk/varnish-mobiletranslate/blob/master/mobile_detect.vcl regex list? That is the same order we use on Mobile_Detect.php it's based on the frequency of devices - this is something researched monthly based on visitor logs.

Any updates to regexes should be double-checked via our UnitTests to prevent regex collisions. ( see: https://github.com/serbanghita/Mobile-Detect/pull/43 )

[willemk]

@lkarsten I've ran your tool on my $20 linode test box and it shows about an 0.1ms (100μs) increase:

With device checking:

root@linode:~/httpttfb/httpttfb# ./measure 8080 Measuring...(dry run) 1 2 3 4 5 done

localhost:8080 (2014-01-22 01:08:17 on linode)

mean: 459.361μs (stdev 58.504μs) median: 450.573μs mean: 309.921μs (stdev 39.811μs) median: 305.879μs mean: 369.633μs (stdev 46.786μs) median: 364.229μs mean: 349.279μs (stdev 39.669μs) median: 347.044μs mean: 386.895μs (stdev 49.983μs) median: 384.037μs

Avg Mean: 374μs

Without device checking: root@linode:~/httpttfb/httpttfb# ./measure 8080 Measuring...(dry run) 1 2 3 4 5 done

localhost:8080 (2014-01-22 01:09:56 on linode)

mean: 307.663μs (stdev 50.453μs) median: 299.464μs mean: 315.561μs (stdev 41.959μs) median: 316.391μs mean: 245.779μs (stdev 37.341μs) median: 226.765μs mean: 245.985μs (stdev 28.984μs) median: 236.468μs mean: 330.259μs (stdev 47.495μs) median: 322.861μs

Avg Mean: 288μs

[willemk]

@serbanghita I've updated the readme file to be a little more user friendly.