EBP-383: Fix runtime cgo panic by changing CCSMP opaque pointer typdefs #77
Conversation
links against CCSMP, it treats all opaque pointers as uintptrs (effectively longs) instead of as void pointers, except for solClient_opaquePtr_pt. I couldn't figure out how to get around solClient_opaquePtr_pt because certain CCSMP functions like solClient_msg_getBinaryAttachmentPtr require a pointer to an opaque pointer. This was all done to avoid the cgo runtime from panicking when it does a runtime check of a C/Go boundary and believes that the 'void *' type that the opaque pointers used to be typedef'd to was an unpinned pointer. As mentioned, the change to the typdef of opaque pointers in solClient.h will prevent the cgo runtime from panicking since it will evaluate the opaque pointers as longs/ints instead of unpinned references. It's unclear whether or not the solClient_opaquePtr_pt will become a problem in the future, but it is actually pointing to data whereas the other solClient_opaque*_pt types are not, which the cgo runtime should be able to discern. The other changes made in this commit were in response to the typdef change, and isolating more memory allocation of configuration structs like the sessionCreateFuncInfo to C in order to mitigate the risk of bad references being accessed by CCSMP.
|
Please mark whether you used Copilot to assist coding in this PR
|
|
@TrentDaniel, also I see a lot of failures on Jenkins - https://jenkins2.solacedev.net/job/solacedev.pubsubplus-go-client/job/EBP-383/3/#showFailuresLink |
included when compiling for the Go API.
|
@TrentDaniel, I still see a lot more failures on Jenkins - https://jenkins2.solacedev.net/job/solacedev.pubsubplus-go-client/job/EBP-383/4/#showFailuresLink . Please can you at least verify that the test failures are not related to the changes in this PR ? |
I've verified that the test failures are not caused by the changes in this PR. Summary:
|
Added compile directive to solClient.h so that when the Go API links against CCSMP, it treats all opaque pointers as uintptrs (effectively longs) instead of as void pointers, except for solClient_opaquePtr_pt. I couldn't figure out how to get around solClient_opaquePtr_pt because certain CCSMP functions like solClient_msg_getBinaryAttachmentPtr require a pointer to an opaque pointer. This was all done to avoid the cgo runtime from panicking when it does a runtime check of a C/Go boundary and believes that the 'void ' type that the opaque pointers used to be typedef'd to was an unpinned pointer. As mentioned, the change to the typdef of opaque pointers in solClient.h will prevent the cgo runtime from panicking since it will evaluate the opaque pointers as longs/ints instead of unpinned references. It's unclear whether or not the solClient_opaquePtr_pt will become a problem in the future, but it is actually pointing to data whereas the other solClient_opaque_pt types are not, which the cgo runtime should be able to discern. The other changes made in this commit were in response to the typdef change, and isolating more memory allocation of configuration structs like the sessionCreateFuncInfo to C in order to mitigate the risk of bad references being accessed by CCSMP.