On Wed, Jun 26, 2002 at 08:12:30AM +0200, Markus Steinmann wrote:
> sorry, I thought my .cvsrc with "diff -u" should be enough.
> Okay, I've now made a "cvs update -Pd" and a
> "cvs diff -u wiretap". Hope this will work.
I just noticed that you used
#pragma pack(2)
in "lanalyzer.h"; unfortunately, that's compiler-specific - not all
compilers support that, and not even all compilers used to build
Ethereal support that.
An alternative would be to declare 2-byte and 4-byte integral quantities
as arrays of 2 and 4 "guint8"s, with additional guint8's added as
padding.
LANalyzer (BTW, that's how it's capitalized on the Novell Web site - see
http://support.novell.com/filefinder/6445/indexb.html
and
http://developer.novell.com/support/sample/tids/lzfwb01/lzfwb01.htm
for example - so that's how it should be capitalized in "file.c"), as
it's a PC program, probably uses little-endian format for multi-byte
integral quantities.
One way to put entries into those 2-byte and 4-byte arrays would to have
a macro that, with shifting and masking, extracts the bytes from a value
and stores them appropriately; for example:
#define PUT32_LE(b, v) { \
(b)[0] = (v) & 0xFF; \
(b)[1] = ((v) >> 8) & 0xFF; \
(b)[2] = ((v) >> 16) & 0xFF; \
(b)[3] = ((v) >> 24) & 0xFF; \
}
which stores its second argument into the array given by the first
argument, as a 32-bit little-endian number. (The equivalent macro for a
16-bit quantity is left as an exercise for the reader.)
That way, you
1) get unaligned quantities in the structures without needing to
use a compiler-specific mechanism;
2) store values into those fields in the correct byte order for
LANalyzer files even on big-endian machines.
Another alternative would be to use "swrite()" to write each field of a
record individually, and to write padding, rather than filling in a
structure and writing the entire structure out in one "swrite()" call.
You'd still have to put stuff out in the right byte order; however, you
could do that by using the "htoles()" and "htolel()" macros from
"wtap-int.h", storing the result into a "guint16" or "guint32" variable,
and writing that variable out.
In addition, "lanalyzer_dump()" is using "guint64"; on most platforms on
which Ethereal is built, the compiler supports 64-bit integral types,
but on at least one platform on which Ethereal has been built (some
versions of SINIX on Siemens
Nixdorf^H^H^H^H^H^H^H^H^H^H^H^H^H^H^HFujitsu Siemens MIPS machines,
using the native compiler), the compiler doesn't support them. That's
why 64-bit arithmetic is done using doubles rather than guint64's in
various files in the wiretap directory.
Also, you should probably replace all occurrences of
REC_TRACE_HEADER with RT_HeaderRegular
REC_CYCLIC_TRACE_HEADER with RT_HeaderCyclic
REC_TRACE_SUMMARY with RT_Summary
REC_TRACE_PACKET_DATA with RT_PacketData
and get rid of the definitions of those 4 REC_ values, so that the same
#defines are used for record types throughout "lanalyzer.c".
You might also want to have "lanalyzer_dump()" return EFBIG rather than
WTAP_ERR_SHORT_WRITE if the record being written to the file would
exceed the maximum size of a LANalyzer file.
