Question

I am writing lexer rules for a custom description language using pyLR1 which shall include time literals like for example:

10h30m     # meaning 10 hours + 30 minutes
5m30s      # meaning 5 minutes + 30 seconds
10h20m15s  # meaning 10 hours + 20 minutes + 15 seconds
15.6s      # meaning 15.6 seconds

The order of specification for hour, minute and second parts shall be fixed to h, m, s. To specify this in detail, I want the following valid combinations hms, hm, h, ms, m and s (with numbers between the different segments of course).
As a bonus the regex should check for decimal (i.e. non-natural) numbers in the segments and only allow these in the segment with least significance.

So I have for all but the last group a number match like:

([0-9]+)

And for the last group even:

([0-9]*\.[0-9]+|[0-9]+(\.[0-9]*)?)  # to allow for .5 and 0.5 and 5.0 and 5

Going through all the combinations of h, m and s a cute little python script gives me the following regex:

(([0-9]*\.[0-9]+|[0-9]+(\.[0-9]*)?)h|([0-9]+)h([0-9]*\.[0-9]+|[0-9]+(\.[0-9]*)?)m|([0-9]+)h([0-9]+)m([0-9]*\.[0-9]+|[0-9]+(\.[0-9]*)?)s|([0-9]*\.[0-9]+|[0-9]+(\.[0-9]*)?)m|([0-9]+)m([0-9]*\.[0-9]+|[0-9]+(\.[0-9]*)?)s|([0-9]*\.[0-9]+|[0-9]+(\.[0-9]*)?)s) 

Obviously, this is a little bit of horror expression. Is there any way to simplify this? The answer must work with pythons re module and I will also accept answers which do not work with pyLR1 if its due to its restricted subset of regular expressions.

Answer 1

You can factorise your regular expression, using the notation h, m, s to denote each of the subregexes, the most basic version is:

h|hm|hms|ms|m|s

which is what you have currently. You can break this into:

(h|hm|hms)|(ms|m)|s

and then pulling out h from the first expression and m from the second we get (using (x|) == x?):

h(m|ms)?|ms?|s

Continuing on we get to

h(ms?)?|ms?|s

which is probably simpler (and probably the simplest).

---

Adding in the regex d to denote decimals (as in \.[0-9]+), this could be written as

h(d|m(d|sd?)?)?|m(d|sd?)?|sd?

(i.e. at each stage optionally have either decimals, or a continuation to the next of h m or s.)

This would result in something like (for just hours and minutes):

[0-9]+((\.[0-9]+)?h|h[0-9]+(\.[0-9]+)?m)|[0-9]+(\.[0-9]+)?m

---

Looking at this, it might not be possible to get into a form ameniable for pyLR1, so doing the parsing with decimals in every spot and then a secondary check might be the best way to do this.