procmailsc - procmail weighted scoring technique
] w^x condition
In addition to the traditional true or false conditions you can specify on a
recipe, you can use a weighted scoring technique to decide if a certain recipe
matches or not. When weighted scoring is used in a recipe, then the final
score for that recipe must be positive for it to match.
A certain condition can contribute to the score if you allocate it a `weight'
) and an `exponent' (x
). You do this by preceding the
condition (on the same line) with:
Whereas both w
are real numbers between -2147483647.0 and
The first time the regular expression is found, it will add w
score. The second time it is found, w*x
will be added. The third time
it is found, w*x*x
will be added. The fourth time w*x*x*x
be added. And so forth.
This can be described by the following concise formula:
n k-1 x - 1
w * Sum x = w * -------
k=1 x - 1
It represents the total added score for this condition if n
Note that the following case distinctions can be made:
- Only the first match will contribute w to the score. Any subsequent
matches are ignored.
- Every match will contribute the same w to the score. The score grows
linearly with the number of matches found.
- Every match will contribute less to the score than the previous one. The
score will asymptotically approach a certain value (see the NOTES
- Every match will contribute more to the score than the previous one. The
score will grow exponentially.
- Can be utilised to favour odd or even number of matches.
If the regular expression is negated (i.e., matches if it isn't found), then
obviously can either be zero or one.
If the program returns an exitcode of EXIT_SUCCESS (=0), then the total added
score will be w
. If it returns any other exitcode (indicating failure),
the total added score will be x
If the exitcode of the program is negated, then, the exitcode will be considered
as if it were a virtual number of matches. Calculation of the added score then
proceeds as if it had been a normal regular expression with
If the length of the actual mail is M
will generate an additional score of:
/ M \
w * | --- |
\ L /
will generate an additional score of:
/ L \
w * | --- |
\ M /
In both cases, if L=M, this will add w to the score. In the former case however,
larger mails will be favoured, in the latter case, smaller mails will be
favoured. Although x can be varied to fine-tune the steepness of the function,
typical usage sets x=1.
You can query the final score of all the conditions on a recipe from the
environment variable $=
. This variable is set every
after procmail has parsed all conditions on a recipe (even if the recipe is
not being executed).
The following recipe will ditch all mails having more than 150 lines in the
body. The first condition contains an empty regular expression which, because
it always matches, is used to give our score a negative offset. The second
condition then matches every line in the mail, and consumes up the previous
negative offset we gave (one point per line). In the end, the score will only
be positive if the mail contained more than 150 lines.
* 1^1 ^.*$
Suppose you have a priority folder which you always read first. The next recipe
picks out the priority mail and files them in this special folder. The first
condition is a regular one, i.e., it doesn't contribute to the score, but
simply has to be satisfied. The other conditions describe things like: john
and claire usually have something important to say, meetings are usually
important, replies are favoured a bit, mails about Elvis (this is merely an
example :-) are favoured (the more he is mentioned, the more the mail is
favoured, but the maximum extra score due to Elvis will be 4000, no matter how
often he is mentioned), lots of quoted lines are disliked, smileys are
appreciated (the score for those will reach a maximum of 3500), those three
people usually don't send interesting mails, the mails should preferably be
small (e.g., 2000 bytes long mails will score -100, 4000 bytes long mails do
-800). As you see, if some of the uninteresting people send mail, then the
mail still has a chance of landing in the priority folder, e.g., if it is
about a meeting, or if it contains at least two smileys.
* 2000^0 ^From:.*(john@home|claire@work)
* 2000^0 ^Subject:.*meeting
* 300^0 ^Subject:.*Re:
* 1000^.75 elvis|presley
* -100^1 ^>
* 350^.9 :-\)
* -500^0 ^From:.*(boss|jane|henry)@work
* -100^3 > 2000
If you are subscribed to a mailinglist, and just would like to read the quality
mails, then the following recipes could do the trick. First we make sure that
the mail is coming from the mailinglist. Then we check if it is from certain
persons of whom we value the opinion, or about a subject we absolutely want to
know everything about. If it is, file it. Otherwise, check if the ratio of
quoted lines to original lines is at most 1:2. If it exceeds that, ditch the
mail. Everything that survived the previous test, is filed.
* 20^1 ^>
* -10^1 ^[^>]
For further examples you should look in the procmailex
(5) man page.
Because this speeds up the search by an order of magnitude, the procmail
internal egrep will always search for the leftmost shortest
unless it is determining what to assign to MATCH
, in which case it
searches the leftmost longest
match. E.g. for the leftmost
match, by itself, the regular expression:
- will always match a zero length string at the same spot.
- will always match one character (except newlines of course).
If, in a length condition, you specify an x
that causes an overflow,
procmail is at the mercy of the pow
(3) function in your mathematical
Floating point numbers in `engineering' format (e.g., 12e5) are not accepted.
As soon as `plus infinity' (2147483647) is reached, any subsequent
conditions will simply be skipped.
As soon as `minus infinity' (-2147483647) is reached, the condition will be
considered as `no match' and the recipe will terminate early.
If in a regular expression weighted formula 0<x<1
, the total added
score for this condition will asymptotically approach:
In order to reach half the maximum value you need
Stephen R. van den Berg
Philip A. Guenther