I did some experimenting, and it appears to me that C (gcc) and DTrace behave the same way, as long as you compare the executions in the same addressing mode.
Gcc defaulted to -m32, which pads after the pointer, because the long long wants to align on a 64 bit boundary. D defaults to 64 bit, so I used the -32 option and got the same results. The whole structure is 24 bytes, but the fields are 4, 4, 4, and 8. Using D's default 64 bit mode, and using the -m64 option in gcc, the structure was still 24 bytes, but the fields were 4, 4, 8, and 8, because the pointer is now 64 bits wide: no padding. Chip > -----Original Message----- > From: dtrace-discuss-boun...@opensolaris.org [mailto:dtrace-discuss- > boun...@opensolaris.org] On Behalf Of Jeff Trawick > Sent: Wednesday, February 11, 2009 9:37 AM > To: dtrace-discuss@opensolaris.org > Subject: Re: [dtrace-discuss] how to get DTrace to use the same > structurefield alignment as C? > > Jeff Trawick wrote: > > Is there a pragma to turn off padding or some other mechanism besides > > dummying up fields to make them char[] to disable padding? > > Here's a better explanation of the issue around my particular example: > > I have a structure which contains some 64-bit integers. > > struct some_api_rec { > int x; > int y; > char *z; > int64_t important_field; > } > > In C, important_field is at offset 0x0C (no padding between z and > important_field). > In D, important_field is at offset 0x10 (4 bytes of padding in front of > important_field to align it for performance, apparently). > > This makes it problematic to use the C structure definition in D > scripts. The C definition can't be changed. > > If I change important_field to > > char important_field[8]; > > the field alignment is fixed, but then important_field will be hard to > use in D scripts. > > _______________________________________________ > dtrace-discuss mailing list > dtrace-discuss@opensolaris.org _______________________________________________ dtrace-discuss mailing list dtrace-discuss@opensolaris.org
