[RFC] get_compiler_info should cache it's results

[RFC] get_compiler_info should cache it's results

[Paul Gilliam]

Hi all,

We are trying to use IBM's xlc compiler with the testsuite and have found 
severial places that break because of differences if flags.  For example, gcc 
uses '-shared' but xlc uses '-qmkshrobj' in order to indicate that a shared 
object is to be produced.

I have written a tcl proc that will make this easyer.  It's kind of like a 
special front-end to 'test_compiler_info', which depends on 
'get_compiler_info' being run first.

I would like to run get_compiler_info from within this new proc, but that 
could result in running it multiple times in a given test.

For this reason, I would like 'get_compiler_info' to cache its resluts by 
simply starting the proc with something like:

if [info exists compiler_info] {return 0}

Does anyone see any problems with this?

-=# Paul #=-

Here is the new proc (so far ;-):  

proc compile_flags { arg1 {arg2 ""} } {
  if {"$arg2" == ""} then {
    set list $arg1
  } else {
    upvar $arg1 lvar
    set list $arg2
  }
  if ![info exists lvar] { set lvar {} }
  get_compiler_info not-used
  foreach {pat flist} $list {
    if {$pat == {default}} {
       return [eval $flist]
    } elseif {[test_compiler_info $pat]} {
      foreach flag $flist {
        set lvar "$lvar additional_flags=$flag"
      }
      set lvar [string trim $lvar]
      return $lvar
    }
  }
  return {}
}

I just realized that this ignores the second argument to 'get_compiler_info' 
which defaults to 'c' but could be 'c++' (and someday may be ada, fortran, 
pascal ...)  I guess my 'compile_flags' would need a 'language' argument to 
pass to 'get_compiler_info', which could cache it's results with an array 
indexed by 'language'.

Re: [RFC] get_compiler_info should cache it's results

[Daniel Jacobowitz]

On Thu, Mar 10, 2005 at 03:57:05PM -0800, Paul Gilliam wrote:
> Hi all,
> 
> We are trying to use IBM's xlc compiler with the testsuite and have found 
> severial places that break because of differences if flags.  For example, gcc 
> uses '-shared' but xlc uses '-qmkshrobj' in order to indicate that a shared 
> object is to be produced.
> 
> I have written a tcl proc that will make this easyer.  It's kind of like a 
> special front-end to 'test_compiler_info', which depends on 
> 'get_compiler_info' being run first.
> 
> I would like to run get_compiler_info from within this new proc, but that 
> could result in running it multiple times in a given test.
> 
> For this reason, I would like 'get_compiler_info' to cache its resluts by 
> simply starting the proc with something like:
> 
> if [info exists compiler_info] {return 0}
> 
> Does anyone see any problems with this?

You need to be slightly more specific - cache it within $board, or
invalidate it somehow.  I don't know precisely when.  GDB's testsuite
isn't very good about this, but you are supposed to be able to run the
testsuite with multiple compilers in one invocation.

I like the idea though!

> 
> -=# Paul #=-
> 
> Here is the new proc (so far ;-):  
> 
> proc compile_flags { arg1 {arg2 ""} } {
>   if {"$arg2" == ""} then {
>     set list $arg1
>   } else {
>     upvar $arg1 lvar
>     set list $arg2
>   }
>   if ![info exists lvar] { set lvar {} }
>   get_compiler_info not-used
>   foreach {pat flist} $list {
>     if {$pat == {default}} {
>        return [eval $flist]
>     } elseif {[test_compiler_info $pat]} {
>       foreach flag $flist {
>         set lvar "$lvar additional_flags=$flag"
>       }
>       set lvar [string trim $lvar]
>       return $lvar
>     }
>   }
>   return {}
> }

I'm not sure about this bit, though.  Could you explain what you want
it to do, and how it would be used?  More informative variable names
might help too.

-- 
Daniel Jacobowitz
CodeSourcery, LLC