Random crashes JPIP in multi-tread environment

[bugbuster]

We have run into random crashes when using JPIP images in a multi-thtread environment. After some tedious debug/logging activities( I am definitely not a multi-thread guru :-) ), I found out where the crash comes from. In the JPIP image destructor (JPIPKAKDataset::~JPIPKAKDataset()) :

• we call the function CPLHTTPCleanup(),
• in turn, this function calls curl_easy_cleanup() on all Curl handles.

In the JPIPKAKDataset destructor, we should only clean our own handle. I proposed a patch which applies to the file jpipkakdataset.cpp only.

However, I do not feel completely satisfied by this. Since I added a direct call to curl_easy_cleanup(), the JPIP plugin has a direct dependency to libCurl.

Shouldn't it better better to do the following :

• add a new function CPLHTTPCleanup2(CURL* myHandle) to cpl_http.cpp,
• use this new function in ~JPIPKAKDataset().

[bugbuster]

patch to jpipkakdataset.cpp

[Even Rouault]

Have a look at ogr/ogrsf_frmts/plscenes/ogrplscenesdataset.cpp for example and OGRPLScenesDataset::~OGRPLScenesDataset() with the CLOSE_PERSISTENT option passed to CPLHTTPFetch()

[bugbuster]

Hi Even,

I have been away from office last week and couldn't answer before. Thanks a lot for your hint. I modified my proposed patch to use the more generic CPLHTTPFetch() I/F. As you can see, I didn't create a new attribute to the JPIPKAKDataset class : instead, I used pszTid, which is declared as an attribute but never used.

[Even Rouault]

In 37431:

JPIPKAK: fix random crashes JPIP in multi-tread environment (fixes #6809)

[Even Rouault]

I've committed a simpler version of your patch in r37431 that doesn't need pszTid (which no longer exists in trunk nor 2.1 branch). I could only verify it compiles but have no way to check runtime behaviour. So please check.

[bugbuster]

It does work with your simplified patch. Thanks