Well, it's one of those things. EDCDSECT does not do a perfect job but it does
a better job than no job at all. You're not using it "in production" so if it
does a 95% job you are 95% ahead. Well, in my experience it is about 95% pretty
much right and about 1% totally wrong, where it gets an offset wrong. Sometimes
it generates such a flipping mess of unions that it is hopeless to read. It
tends to get unions wrong. How do I say this? Something that logically should
have been rendered
union x {int a; int b;};
union y {int c; int d;};
it instead renders as
union z {int a; int c;};
union w {int b; int d;};
Yes, of course, assembler is weakly typed, and even there, people misuse what
typing there is. EDCDSECT has some options, for example, how do you want to
deal with
FLAG DS X
INPUT EQU X'80'
OPEN EQU X'40'
etc.
Do you want INPUT to be a one-bit integer or a #define mask suitable for use
with logical or? EDCDSECT gives you a choice.
You have a choice whether to lowercase the assembler symbols (because old
assembler is typically all uppercase and C culture is lowercase).
Yes, there would be some challenges going the other direction: Foo and foo are
different C symbols but the same assembler symbol (but do you use both in the
same struct???). C struct symbols typically have local scope; all assembler
symbols have global scope. You could deal with the latter by letting the user
specify a prefix or by rendering
struct Foo { int Bar; };
in assembler as
Foo_Bar DS F
Charles
-----Original Message-----
From: IBM Mainframe Discussion List [mailto:[email protected]] On Behalf
Of Paul Gilmartin
Sent: Saturday, March 19, 2016 2:26 PM
To: [email protected]
Subject: Re: C struct to DSECT utility?
On Sat, 19 Mar 2016 12:22:44 -0700, Charles Mills wrote:
>Off the wall question: IBM ships EDCDSECT, which will convert an
>assembler DSECT (actually, the ADATA from assembling a DSECT) to a C struct.
>
I had heard of such utilities from ISVs; I had been unaware of this from IBM.
Like Samuel Johnson's politically-incorrect dog, it surprises me that this is
done at all. Some challenges:
o Indiscriminate use of ORG, not always readily rendered as "union".
o Weakness of assembler typing. For example, should:
A DC A(SYMBOL)
be rendered as which of:
char *A = &SYMBOL; /* or: */
long A = SYMBOL;
One can take a guess according to whether SYMBOL is relocatable or absolute and
be right some of the time.
>Is there anything, or combination of things, that will go the other
>direction? Convert a C struct into a DSECT?
>
Backward? Likewise challenges:
o Similar member names in different structs. Qualify them?
o References to typedefs and structure tags from earlier headers.
o Judicious reliance on case-sensitivity. Ok; I'll grant that I'm a partisan\
and I'd not be surprised to encounter nor loath to use in a C program
such as:
char *Auml = "Ä"
char *auml = "ä"
For IBM-MAIN subscribe / signoff / archive access instructions, send email to
[email protected] with the message: INFO IBM-MAIN
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN
