[WMS Server] Crash when using wms_style* metadata

[jmckenna@…]

I am documenting the changes implemented in bug#944.

When trying the new wms_style metadata for LegendURL (with mapserv 4.4-beta3), I
cannot generate a warning, and if I try to correctly specify the new metadata my
mapserv.exe crashes.

here is an example layer that should work:

LAYER
  NAME land_fn
  METADATA
    "wms_title"  "Foreign Lands"
    "wms_style" "test" 
    "wms_style_test_legendurl_width" "72"    
    "wms_style_test_legendurl_legend" "72"  
    "wms_style_test_legendurl_format" "image/gif"
    "wms_style_test_legendurl_href" "http://wwww.rrr.com"
  END
  ...

[jmckenna@…]

mapserv crashes on a GetCapabilities request

[dmorissette]

This happens because of poor management of the pszMetadataName in msDumpLayer()
when writing the legendURl coming from metadata. Cleaning this up now.

[dmorissette]

Important: to reproduce this, use VERSION=>1.1.0, e.g.

http://sb1/cgi-bin/mapserv_44?map=/path/to/my.map&version=1.1.1&request=getcapabilities

[dmorissette]

Fixed in 4.4.0

The problem was a stray msFree(pszMetadataName) in msDumpLayer() in mapwms.c.
Removing the extra msFree() fixed the crash. I also removed all other msFree()
calls for that variable and kept a single one at the end of the function.

[dmorissette]

BTW Jeff, in case you didn't notice already, in your example there was a typo in
the height metadata, it should be:
    "wms_style_test_legendurl_height" "72"  

[jmckenna@…]

right, thanks

[jmckenna@…]

verified.

however the functionality implemented in bug#944 for legendURL is this:

a warning is displayed only if both "wms_style" and
"wms_style_test_legendurl_href" are specified...i would think the desired
functionality would be that if a user is missing ANY of the required metadata
(width, height, format, href) for legendURL a warning should be displayed.  

[dmorissette]

Producing a warning in all cases would just increase the complexity of code with
little benefit. As things are currently implemented, it's the legendurl_href
that triggers the whole legendurl stuff and that makes sense to me since that's
the most important piece of information and the least likely to be forgotten
(without an HREF, you can't even imagine having a LegendURL). 

What you're asking is that the presence of any of the metadatas could trigger a
cross-check of all other metadatas? You could always file an enhancement bug
about this if you think this would add value, but I believe that not
unnecessarily increasing code complexity should also be taken into account when
deciding what to do about this.

[jmckenna@…]

ok agreed.  I just threw that out there for you and Tom just in case.