Experimental Physics and
Industrial Control System

Andrew Johnson <[email protected]> · Mon, 12 Mar 2007 11:20:41 -0500

Ralph Lange wrote:



just found out (compiling streamDevice that includes a device support

for the calcout) that the generated header file calcoutRecord.h uses

definitions from postfix.h, but does not include it.

I know this is not illegal, but it's not very nice, too.



The postfix.h requirement is new in R3.14.9 - I did document it in the

release notes, but I agree that it's not very nice.



As calcoutRecord.h gets generated from dbd - is there an easy way to add

the #include directive or should we live with calcoutRecord.h being not

nice?



There was supposed to be -- I added support ("Cdefs") for this kind of

thing to dbStaticLib before R3.14.0alpha1 came out, but we've never used

it in anger and unfortunately it doesn't work for this particular case.

 I suspect VDCT doesn't understand the Cdefs syntax either, so it may

barf when it sees a Cdefs line. This is the DBD syntax:


Index: calcoutRecord.dbd
===================================================================

RCS file:

/net/phoebus/epicsmgr/cvsroot/epics/base/src/rec/calcoutRecord.dbd,v

retrieving revision 1.5.2.2
diff -u -b -r1.5.2.2 calcoutRecord.dbd

--- calcoutRecord.dbd   3 Oct 2006 20:22:23 -0000       1.5.2.2
+++ calcoutRecord.dbd   12 Mar 2007 15:17:07 -0000
@@ -30,6 +30,7 @@
        choice(calcoutINAV_CON,"Constant")
 }
 recordtype(calcout) {
+       %#include "postfix.h"
        include "dbCommon.dbd"
        field(VERS,DBF_DOUBLE) {
                prompt("Code Version")


Any valid C code can follow a % sign (which must be inside a recordtype

definition) and will be added to the generated *Record.h file.

Unfortunately though the above "#include" statement gets added to the

calcoutRecord.h file *below* the struct calcoutRecord definition, which

is too late for postfix.h. I can fix that, but obviously not until the

next release.



In fact looking at it, I can clean up the dbStaticLib code slightly

using this technique - I can move all of the #include statements that

get put into the Record.h files from the dbToRecordTypeH.c program into

dbCommon.dbd, so this makes sense for the next release (but for now all

calcout device support will have to do the #include "postfix.h" itself).



The original idea for Cdefs was to allow each recordtype to define its

own DSET structure for the device supports to use, rather than each

support defining its own. I eventually concluded though that we're too

far gone to try and fix that particular design issue until we re-do the

whole devSup interface.


- Andrew
--
The right to be heard does not automatically include
the right to be taken seriously. -- Hubert H. Humphrey


Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System