*! mcompr1.ado          (sg28: STB-22)
program define mcompr1
    version 3.1
    local options "GENerate(string) Cutoff(real .05) ID(string) Default(string) LAbel"
    local varlist "req ex max(1)"
    parse "`*'"
    parse "`varlist'", parse(" ")
    if "`default'"=="" { local default default }
    local numcat = $S_2 + ("$S_4"!="nocons")
    tempvar bb vv ll vv2
    qui gen `bb' = .
    if "`id'"=="" { tempname id }
    qui gen str28 `id' = ""
    local i 1
    local ii 1
    if ("$S_4"!="nocons") {
        qui replace `bb' = 0 in 1
        if "`id'"!="" { qui replace `id' = "`default'" in 1 }
        local i 2
    }
    while `i' <= `numcat' {
        local nam : word `ii' of $S_3
        qui replace `bb' = _b[`nam'] in `i'
        if ("`label'"!="") { local nam : variable label `nam' }
        qui replace `id' = "`nam'" in `i'
        local i = `i' + 1
        local ii = `ii' + 1
    }
    qui compress `id'
    if "`generat'"=="" { tempname generat }
    qui gen str20 `generat' = ""
    gen long `vv' = _n
    sort `bb' `vv'
    /* we would like vv2 to be such that when sorted by vv,
       vv2 would point to the list of contrasts in order.
       In other words, vv2[i] would identify the i-th item
       in the list.

       This would happen if we could compute vv2[vv] = _n
       vv  bb  rank vv2        vv, when sorted by bb
       1   1.2  3     2        2
       2   0.6  1     3        3
       3   0.8  2     1        1

    an easy solution would be to save vv in the present sorted order and
    merge it back after resorting by vv!
    */

    tempfile USER VV
    cap drop _merge
    qui save `USER'
    qui keep `vv'
    rename `vv' `vv2'
    cap save `VV', replace
    qui use `USER', clear
    sort `vv'
    qui merge using `VV'
    drop _merge

    local i 1
    local g 1
    local top 0
    while `i' <= `numcat' {
        local j = `i'+1
        local cont 1
        while (`j' <= `numcat') & `cont' {
            local ii = `i'
            while `ii'<`j' {
                local i1 = min(`vv2'[`ii'],`vv2'[`j'])
                local j1 = max(`vv2'[`ii'],`vv2'[`j'])
                local pat = (`j1'-1)*(`j1'-2)/2 + `i1'
                local p = `1'[`pat']
                if (`p' <= `cutoff') { local cont 0 }
                local ii = `ii' + 1
            }
            if `cont' { local j = `j' + 1 }
        }
        * At this point, we have identified a group running from i to j-1
        * such that all of these objects are within the cutoff P of each
        * other.  We dont yet know if they were included in a similar
        * group with a lower value of i.  the variable "top" will tell us
        if (`top'<`j'-1)  {
            local top = `j'-1
            local ii  `i'
            local sym = substr("ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789abcdefghijklmnopqrstuvwyz",mod(`g'-1,62)+1,1)
            di "Group `sym'"
            while `ii'<`j' {
                local k = `vv2'[`ii']
                di " " `id'[`k']
                local ks = `generat'[`k']
                qui replace `generat' = substr("`ks'`sym'",1,20) in `k'
                local ii = `ii' + 1
            }
            local g = `g' + 1
        }
        local i = `i' + 1
    }
    qui compress `generat'
end