On Wed, Mar 09, 2016 at 11:18:28AM +1100, Ben Elliston wrote:
> On Tue, Mar 08, 2016 at 02:29:10PM +0100, Dominik Vogt wrote:
>
> > However, multiple dg-do don't seem to work:
> >
> > dg-do run { condition }
> > dg-do assemble { ! condition }
>
> There are plenty of examples of this in the testsuite. You may wish to
> investigate further:
>
> $ grep -r dg-do . | awk -F: '{print $1}' | uniq -d
>
> For example, the top of gcc.target/powerpc/altivec-3.c says:
>
> /* { dg-do run { target { powerpc*-*-* && vmx_hw } } } */
> /* { dg-do compile { target { powerpc*-*-* && { ! vmx_hw } } } } */
>
> So compile awyas on powerpc-*-*, but only run if the taget has the
> right hardware. It should certainly work. If it doesn't, please send
> a patch!
The problem is really that if a condition deselects a test
(dg-process-target returns 'N'), and an earlier dg-do already
selected the test, the first word of $dg-what still gets replaced
by the new (deselected) action. E.g.
* $dg-what is "compile {} P" at first (default value)
* becomes "run S P" after evaluating the selected "dg-do run"
* becomes "assemble S P" after ervaluating deselected "dg-do" assemble.
The attached patch fixes the problem for me by not changing the
action if a deselected dg-do is encountered. Note that my
knowledge of Tcl is about zero, so there may be any number of bugs
in that patch.
Ciao
Dominik ^_^ ^_^
--
Dominik Vogt
IBM Germany
--- dg.exp.orig 2013-08-03 09:27:13.000000000 +0200
+++ dg.exp 2016-03-09 13:45:41.000000000 +0100
@@ -313,6 +313,7 @@
return
}
+ set doaction [lindex $args 1]
set selected [lindex ${do-what} 1] ;# selected? (""/S/N)
set expected [lindex ${do-what} 2] ;# expected to pass/fail (P/F)
@@ -330,6 +331,8 @@
# are OR'd together).
if { $selected != "S" } {
set selected "N"
+ } else {
+ set doaction [lindex ${do-what} 0]
}
}
"F" { set expected "F" }
@@ -355,7 +358,7 @@
error "[lindex $args 0]: syntax error"
}
}
- set do-what [list [lindex $args 1] $selected $expected]
+ set do-what [list $doaction $selected $expected]
}
proc dg-error { args } {
_______________________________________________
DejaGnu mailing list
DejaGnu@gnu.org
https://lists.gnu.org/mailman/listinfo/dejagnu
