[ 
https://issues.apache.org/jira/browse/THRIFT-4781?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17100817#comment-17100817
 ] 

Xusheng Zhang commented on THRIFT-4781:
---------------------------------------

First, I want to thanks [~emmenlau] and [~andrejn] for your reply.

[~andrejn]: A few days ago, I haved tried to remove files from the 
libthrift.vcxproj file and build successful. But my client of project have some 
strange errors while building. From the picture you can see 
thrift::concurrency::xxxx which all point to the concurrency(I'm sorry that my 
VS2019 is in Chinese so the pic have some Chinese). That's why I suspect the 
thrift cpp library build have potential errors and tried it again.

I don't try cmake, maybe I will try it in a few days.

!image-2020-05-06-21-37-46-809.png!

> C++ clients crash when exceptions are typedefed in the IDL
> ----------------------------------------------------------
>
>                 Key: THRIFT-4781
>                 URL: https://issues.apache.org/jira/browse/THRIFT-4781
>             Project: Thrift
>          Issue Type: Bug
>          Components: C++ - Compiler
>    Affects Versions: 0.11.0, 0.12.0
>            Reporter: Andrej Nazarov
>            Priority: Major
>         Attachments: image-2020-05-06-20-15-44-986.png, 
> image-2020-05-06-21-37-46-809.png
>
>
> If exceptions are typedefed in the IDL, they're generated as pointers in Cpp. 
> This causes a runtime crash (memory access violation) on the C++ client-side 
> when a server sends that exception and the client tries to read it. Example 
> follows:
> {code:java|title=service.thrift}
> namespace * thrifttest.service
> include "errors.thrift"
> typedef errors.FooError FooError
> service FooBarService
> {
>       string getFooString(1: i32 stringLength) throws (1: FooError e);
>       string getBarString(1: i32 stringLength) throws (1: errors.BarError e);
> }
> {code}
> {code:java|title=errors.thrift}
> namespace * thrifttest.errors
> exception FooError {
>   1: string message
> }
> exception BarError {
>   1: string message
> }
> {code}
> {code:java|title=FooBarService.h}
> class FooBarService_getFooString_presult {
>  public:
>   virtual ~FooBarService_getFooString_presult() throw();
>   std::string* success;
>   FooError* e; //note pointer declaration of the exception field
> // snip...
> class FooBarService_getBarString_presult {
>  public:
>   virtual ~FooBarService_getBarString_presult() throw();
>   std::string* success;
>    ::thrifttest::errors::BarError e; //note different declaration of the 
> exception field
> //snip
> {code}
> {code:java|title=FooBarService.cpp}
> uint32_t 
> FooBarService_getFooString_presult::read(::apache::thrift::protocol::TProtocol*
>  iprot) {
> // snip...
>   while (true)
>   {
> // snip...
>     switch (fid)
>     {
> // snip...
>       case 1:
>         if (ftype == ::apache::thrift::protocol::T_STRUCT) {
>           xfer += (*(this->e)).read(iprot); // <-- this line causes access 
> violation crash because the pointer is not initialized
>           this->__isset.e = true;
>        // snip...
> uint32_t 
> FooBarService_getBarString_presult::read(::apache::thrift::protocol::TProtocol*
>  iprot) {
> // snip...
>   while (true)
>   {
> // snip...
>     switch (fid)
>     {
> // snip...
>       case 1:
>         if (ftype == ::apache::thrift::protocol::T_STRUCT) {
>           xfer += this->e.read(iprot); //<-- this gets read OK.
>           this->__isset.e = true;
> //snip
> {code}
> This happens regardless of server language (reproducible if server throwing 
> the exceptions is Java, Python or C++)
>  I guess this logic in 
> [t_cpp_generator.cc:1104|https://github.com/apache/thrift/blob/0.11.0/compiler/cpp/src/thrift/generate/t_cpp_generator.cc#L1104]
>  gets deceived in case of typedefed exceptions:
> {code:java|title=t_cpp_generator.cc}
> (pointers && !(*m_iter)->get_type()->is_xception()),
> {code}
> I'm no Thrift compiler expert, but I assume there is a reason why you don't 
> want exceptions to be declared as pointers. Yet in this case they clearly are.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to