Question

I am chasing an exception which is thrown from a part of code which is added by the compiler after each call to new. It’s the standard C++ new, which should get some memory from the heap and call the constructor of the class.

We are running VxWorks 5.5.1 with GCC 2.95 (or 2.96 not sure about that) on a SH4 processor. Compiled within SNiFF+ 4.1 Patch 1.

The C++ codes looks like this.

CBlocksFile* pBlockFile = new CBlocksFile(szHeaderFile, szDataFile);

And the generated assembler code has a terminate/delete/throw handling after the call to new. This pattern seems to be applied on all calls to new.

// call to "new"
c4d6000  d14d        mov.l      @(0x134,pc),r1 (= 0x0c06c1e0 = ___builtin_new)
c4d6002  410b        jsr        @r1
c4d6004  e414       (mov        #20,r4)
...
// compiler generates throw path address
c4d6022  d246        mov.l      @(0x118,pc),r2 (= 0x0c4d6034)
...
// and pushes it to the stack
c4d602c  1121        mov.l      r2,@(4,r1)
...
c4d6030  a002        bra        +4       (==> 0x0c4d6038 : GOOD_PATH)
...
// throw path (there is no visible jump to this address)
c4d6034  a088        bra        +272       (==> 0x0c4d6148 : THROW_PATH)
...
GOOD_PATH:
...
// call to constructor
c4d6058  d139        mov.l      @(0xe4,pc),r1 (= 0x0c4d1730 T ___Q211CBlocksFilePCcT1bUcl)
...
c4d6060  410b        jsr        @r1
...
// normal path
return

THROW_PATH:
...
// same pattern again, compiler generates terminate path address
c4d6164  d22f        mov.l      @(0xbc,pc),r2 (= 0x0c4d6172)
...
// and pushes it to the stack
c4d616a  1121        mov.l      r2,@(4,r1)
...
c4d616e  a002        bra        +4       (==> 0x0c4d6176 : NO_TERMINATE)
...
c4d6172  a039        bra        +114       (==> 0x0c4d61e8 : TERMINATE_A)
...
NO_TERMINATE:
...
// delete handling
if ( ?? )
c4d617c  2118        tst        r1,r1
c4d617e  8d04        bt/s       +8       (==> 0x0c4d618a)
...
{
    delete ??
...
c4d6184  d128        mov.l      @(0xa0,pc),r1 (= 0x0c06be20 = ___builtin_delete)
c4d6186  410b        jsr        @r1
...
}
c4d618a  9044        mov.w      @(0x88,pc),r0 (= 0x0000028c)
c4d618c  02ee        mov.l      @(r0,r14),r2
c4d618e  5121        mov.l      @(4,r2),r1
c4d6190  6112        mov.l      @r1,r1
c4d6192  1211        mov.l      r1,@(4,r2)
c4d6194  d125        mov.l      @(0x94,pc),r1 (= 0x0c06a880 = ___sjthrow)
c4d6196  410b        jsr        @r1

For what is this code snipped useful? It does not look like the no memory condition, because the throw is not within the new function.

Why is the throw called? And from whom? There is twice this pattern of putting a code address onto the stack, which might be used for execution later on.

Thanks

Answer 1

A guess would be that this pattern is applied because it is a requirement of c++ that if the constructor throws, the memory for the object should be freed. So if an exception is thrown from within the constructor, the newly allocated object gets deleted.

You could always try to explicitly throw something from the constructor of your object to see which path execution takes to verify whether or not this answer is useful.