Wireshark-dev: Re: [Wireshark-dev] Display Filter Fields typos, missing fields, etc.

On Aug 22, 2007, at 12:39 PM, Maynard, Chris wrote:

I was looking at the display filter fields recently and noticed that
there seem to be some display filter field naming mistakes/typos.
Basically, there are a lot of naming inconsistencies, and in at least
one case, no name at all. Here are a just a few of the dissectors that
seem affected:

1.	There are many fields (I count 51 of them) that seem to me that
they should be prefixed with "wlan.".  For example, fixed.chanwidth,
fixed.qosinfo.ap, fixed.qosinfo.ap.edcaupdate, ...

Yes, those are probably typoes.

2.	hf_tag_measure_rpi_histogram_report is missing a display filter
name, i.e.,
		{"Receive Power Indicator (RPI) Histogram Report", "",

That one probably is a typo/mindo, too.

		There appear to be 4 different prefixes used: "slow.",
"lacp.", "marker.", and "oam.".  Should they all be prefixed with

Possibly. The Link Aggregation Control Protocol, the Link Aggregation Marker Protocol, and the Operation, Administration and Maintenance protocol are all arguably separate protocols; however, they don't have independent Ethertypes, they all use the "slow protocols" Ethertype, with the "slow protocols" header having a 1-byte subtype field specifying which particular slow protocol the packet contains. As such, perhaps the dissector should dissect only the subtype as the "slow" protocol, with separate dissectors for all the individual slow protocols.

		Two fields are prefixed with "aim." and one is prefixed
with "admin.".  Shouldn't all 3 be prefixed with, "aim.admin."?

Or, given that the protocol is "aim_admin", perhaps they should all be prefixed by "aim_admin.".

1.	"sm_a.rr.pow_cmd_pow" should probably be "gsm_a.rr.pow_cmd_pow".

Probably just a typo.

There are a lot more, but for brevity, I haven't listed them all here.
Is reporting this on the developer's list sufficient, or should I open a
bug for all/some of them?

I'd open bugs for each dissector (so we don't have to fix them all at once to close the bug).

Perhaps there should be a more strict naming convention?

Or a more clearly stated naming convention. The general convention is that a field name's first component should be the name of the protocol to which it belongs.

Is there a way to automate the names so typos, missing fields, etc. can be avoided?

With the existing scheme in which the field list is constructed by hand, about all you could do would be to use macros and compile-time concatenation, or do string processing at registration time; the latter might slow down registration a bit, and increase memory use.

Dissectors generated automatically could probably do better. There are a number of reasons why, in general, it's a Good Thing to generate dissectors from specification languages, and that's one of them.

Registration-time checking of the fields could also help - or, perhaps, instead of doing the checking all the time, just do it at build time, with a special option to tshark, or a special checker program (Did somebody already write something such as that? I vaguely remember something of that sort.).