CVS difference for ais/ai-00041.txt

Differences between 1.7 and version 1.8
Log of other versions for file ais/ai-00041.txt

--- ais/ai-00041.txt	2000/03/09 03:09:40	1.7
+++ ais/ai-00041.txt	2000/04/11 19:36:33	1.8
@@ -839,3 +839,68 @@
 reflect this and other changes.
 
 ****************************************************************
+
+From: Tucker Taft
+Sent: Monday, March 13, 2000 1:24 PM
+
+Erhard Ploedereder wrote:
+>
+> The AI talks about overriding pragmas on the instances. How do I override
+> Inline, Export, Import ?  I'd love to, especially for Inline. :-)
+>
+> More seriously, the only pragma that's overridable is Convention. In that
+> sense, the proposed wording is still correct, but a bit sweeping.  Should it
+> be changed ?
+
+We could add some "weasel" words to make it clear that not all pragmas
+have mechanisms for overriding their effect.
+
+****************************************************************
+
+From: Dan Eilers
+Sent: Monday, March 27, 2000 7:20 PM
+
+I'd like to second this request regarding overriding pragma inline,
+even if it wasn't completely serious.
+
+We support a pragma "noinline" that prevents both explicit and implicit
+inlining.  This is useful for a variety of purposes, such as preventing
+inlining on subprograms with machine_code, subprograms that are
+called infrequently, and subprograms that the user wishes to be able
+to set a breakpoint on in the debugger.  It is also useful when
+investigating possible code generation anomalies.
+
+****************************************************************
+
+From: Robert Dewar
+Sent: Monday, March 27, 2000 7:27 PM
+
+Why are subprograms with machine code an issue? Those seem among the
+most important to inline ???
+
+****************************************************************
+
+From: Dan Eilers
+Sent: Monday, March 27, 2000 7:34 PM
+
+Normally yes, but sometimes not.
+For example, a user might reference registers used for parameter passing,
+which would be nonsense if inlined.
+
+****************************************************************
+
+From: Robert Dewar
+Sent: Monday, March 27, 2000 7:40 PM
+
+But surely such references should be symbolic, using the parameter names,
+rather than using absolute registers???
+
+****************************************************************
+
+From: Dan Eilers
+Sent: Monday, March 27, 2000 8:25 PM
+
+Yes, generally they should be symbolic.  But sometimes a user will
+reference absolute registers anyway, maybe for good reasons and maybe not.
+
+****************************************************************

Questions? Ask the ACAA Technical Agent