NetNews Usenet Archive 1992 #30/NN_1992_30.iso/spool/comp/lang/fortran/4728

Text File | 1992-12-15 | 3.5 KB | 80 lines

Path: sparky!uunet!timbuk.cray.com!shamash!ems!ems.cdc.com!mstemper
From: [e-mail] (Michael Stemper)
Newsgroups: comp.lang.fortran
Subject: Re: Need help with named common blocks
Message-ID: <31152@nntp_server.ems.cdc.com>
Date: 15 Dec 92 18:35:03 GMT
References: <[e-mail]> <[e-mail]>
Sender: [e-mail]
Reply-To: [e-mail]
Organization: Empros Systems International, a division of Ceridian
Lines: 66
Nntp-Posting-Host: kirk.ems.cdc.com

In article <[e-mail]>, [e-mail] writes:
>In article <[e-mail]>, [e-mail] (Michael Lanham) writes:

>> I have several program src files.  In the file with the BLOCK DATA segment,
>> the COMMON block is named and shaped as follows:

>>       COMMON/SHARE1/TABLE(1770),NOPU(54),FLHS(54),UNUSE0(122),
...
>>      -           LISTX(61),INTERX(70),SYNTAB(2,26),CHARX(26),TAG(50),
>>      -           (50)

>> Now I found an additional SHARE1 that is different.
               ********************

>>       COMMON/SHARE1/NSVT(1000),OSVT(1000),
...
>>      -             OUTTAB(50,4),SUPP(100,2),SYMBL(2),UNUSE2(18)

>> Is an accurate way of thinking of SHARE1 is the first declaration plus the
>> second concatenated to the end of the second?  I need to know this so I can
>         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

>    No.  In FORTRAN, the named common is a single storage area.  A mis-feature
>of FORTRAN is that common blocks may be declared differently in different
>program sections, but each still refers to the same storage area.  Thus, in
>the example above, NSVT(1) is the same storage location as TABLE(1), etc.

Assuming that Mr. Lanham's words "an additional SHARE1" mean that the
module in question had both declarations, then his guess was right, the
two definitions are concatenated. However, only this module has access
to the additional items (unless somebdoy did something really ugly, like
accessing TAG out of bounds to get at the extras).

If the module only had this new declaration, then your reasoning follows.

>of type CHARACTER*1 (or BYTE or INTEGER*1 or LOGICAL*1)...  The point is
>that most compilers allow the total length of a common to be different in
>different program sections.  Some will (nicely) take the longest declared
>length as the total, others (not so nice) will take the _last_ declared
>length.

Assuming each module is separately compiled, there is absolutely nothing
that a compiler can do to prevent different modules from having different
lengths for the same COMMON block. However, as you point out, a common
block is a single storage area. The loader must figure out how much space
to allocate for it.

The strategy that I am most familiar with is the first time a loader
encounters a COMMON block, it will reserve as much space as the module
referencing it requests. If modules loaded later think that the block
is shorter, there is no problem. If one of them thinks it is longer,
you are S.O.L. If the loader is friendly, it will give you a fatal
error message. Otherwise, it won't detect the problem and you will
end up overwriting the beginning of another COMMON block (or worse
yet, code!).

>    In any case, you may want to investigate how /SHARE1/ is being used in
>the odd-man-out routine to be sure it's not an out-and-out error...

ABSOLUTELY!!!

--
#include <Standard_Disclaimer.h>
Michael F. Stemper
Power Systems Consultant
[e-mail]