Re: st: RE: Obtain the list of variables used in an if condition

[Austin Nichols <austinnichols@gmail.com>]

Sergiy et al.--
I wouldn't worry about macros since they will be evaluated, but I am
probably missing some other problems. Here was my imperfect first cut:

clear all
prog shvars
 syntax [if]
 tempname x
 loc if:subinstr loc if "c(" "`x'", all
 loc if:subinstr loc if "r(" "`x'", all
 loc if:subinstr loc if "e(" "`x'", all
 tokenize `"`if'"', parse(" !~^+-*/=<>]()&|")
 while `"`1'"'!="" {
  cap unab v: `1'
  if _rc==0 loc varlist "`varlist' `v'"
  mac sh
  }
 di "`varlist'"
 end
sysuse auto
su mpg
g meanvar=1
shvars if mpg==r(mean)&(di==_pi|gear_r==c(pi))

Here is one headache:

. shvars if mpg==r(mean)&(di==_pi|gear_r==c(pi))&floor(f)==0
 mpg displacement gear_ratio

. shvars if mpg==r(mean)&(di==_pi|gear_r==c(pi))&trunc(f)==0
 mpg displacement gear_ratio

. shvars if mpg==r(mean)&(di==_pi|gear_r==c(pi))&reverse(make)==""
 mpg displacement gear_ratio

I.e. it is not good enough to strip out r( and c( and e( since these
can be the end of a function.

Here is one fix, but still not complete (no mention of factor variables yet):

clear all
prog shvars
 syntax [if]
 tempname x
 foreach h in c e r {
  foreach a in `c(alpha)' {
   loc if:subinstr loc if "`a'`h'(" "`a'x(", all
   }
  loc if:subinstr loc if "`h'(" "`x'", all
  }
 tokenize `"`if'"', parse(" !~^+-*/=<>]()&|")
 while `"`1'"'!="" {
  cap unab v: `1'
  if _rc==0 loc varlist "`varlist' `v'"
  mac sh
  }
 di "`varlist'"
 end
sysuse auto
su mpg
g meanvar=1
shvars if mpg==r(mean)&(di==_pi|gear_r==c(pi))
shvars if mpg==r(mean)&(di==_pi|gear_r==c(pi))&floor(f)==0



On Wed, Aug 12, 2009 at 4:48 PM, Jeph Herrin<junk@spandrel.net> wrote:
> Another consideration is that an -if- condition can
> refer to a scalar
>
> ********************
> program foo
>        syntax [if]
>        di "`if'"
> end
> ********************
> . clear, all
> . scalar cond=1
> . foo if cond==1
> if cond==1
>
> not to mention locals and globals.
>
> Good luck!
>
> Jeph
>
>
>
>
> Nick Cox wrote:
>>
>> I don't know a neat way to do this. Among other problems, a complete
>> solution would need to consider that
>> 1. An -if- condition might contain temporary variable names, too. Your
>> logic seems to hold despite that.
>> 2. An -if- condition could refer to r(), e(), c() and Stata matrices.
>> Your logic might need some modification there.
>> Nick n.j.cox@durham.ac.uk
>> Sergiy Radyakin
>>
>> how do I get the list of variables used in an if-condition passed to
>> my Stata program?
>> obviously all sorts of logical expressions and functions are permitted
>> in the if-condition.
>>
>>
>> program foobar
>>    syntax [if]
>>
>>    // ... determine which vars are used in the if-condition and put
>> them in var_list ...
>>    // ... compute something ...
>>
>> end
>>
>>
>> I am looking for an elegant solution, the one that would not involve,
>> say, dropping vars one
>> by one and checking if the expression is still computable.
>>
>> One solution is to replace all the ! ~ ^ + - * / = < > ] and )  to
>> spaces, then look at the tokens left.
>> Those with ( immediately following are function names, those with [ or
>> a space immidetely
>> following are variables. (and we can use -capture confirm variable- to
>> check that).
>>
>> Is there any problem with such an approach? (e.g. factor variables in
>> Stata 11?)
>> Is there a better (or a standard/existing) approach to the problem?

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/