Re: Is this the bug of wmem_tree_lookup32_array_le()?

["qiangxiong.huang via Wireshark-dev" <wireshark-dev () wireshark org>]

John Thacker,


Thank you for your information. I may try to add xxx_insert64/lookup64().


------------------ Original ------------------
From:                                                                                                                   
     "Developer support list for Wireshark"                                                                             
       <johnthacker () gmail com&gt;;
Date:&nbsp;Tue, Oct 19, 2021 10:42 PM
To:&nbsp;"Developer support list for Wireshark"<wireshark-dev () wireshark org&gt;;

Subject:&nbsp;Re: [Wireshark-dev] Is this the bug of wmem_tree_lookup32_array_le()?



On Sun, Oct 17, 2021 at 5:41 AM qiangxiong.huang via Wireshark-dev <wireshark-dev () wireshark org&gt; wrote:



Who knows that the current behavior of the wmem_tree_lookup32_array_le() is designed in this way or it is just a bug?
If I want to fix the issue #17633, should I modify this wmem_tree_lookup32_array_le() directly or write a new function 
like wmem_tree_lookup32_array_le_key_as_big_num() (or other name you recommend)?



While I fully understand why you would want a lookup that would be a lexicographical sort with the order of the 32 bit 
integers in the key being rank (in order to handle a 64 bit integer as two 32 bit integers), the current dissectors 
that use the API depend on the current behavior and would break, e.g. 
https://gitlab.com/wireshark/wireshark/-/blob/master/epan/dissectors/packet-dns.c#L4184-L4185 and 
https://gitlab.com/wireshark/wireshark/-/blob/master/epan/dissectors/packet-usb.c#L3560-L3574


Generally these dissectors have several independent keys, and what they want to match is an exact match on some keys 
(transaction ID, for example) and a less than match on others (e.g., frame number.) They do the less than or equal 
match and then check for strict equality on the keys that need that. Handling the lookup like a lexicographical sort 
would give incorrect results for those dissectors; the current behavior is as intended.



So if you're going to do it, you need a new API function. But perhaps what you really want is to implement something 
like wmem_tree_lookup64 that supports 64 bit integers directly rather than expecting the callers to separate into the 
upper and lower 32 bits. It is possible to do lookups of other types (there's a string lookup using a string comparison 
function, which might be worth looking at since the method for 32 bit integers calls the GUINT_TO_POINTER macros and is 
guaranteed to work for 64 bit integers.)



John Thacker

___________________________________________________________________________
Sent via:    Wireshark-dev mailing list <wireshark-dev () wireshark org>
Archives:    https://www.wireshark.org/lists/wireshark-dev
Unsubscribe: https://www.wireshark.org/mailman/options/wireshark-dev
             mailto:wireshark-dev-request () wireshark org?subject=unsubscribe