https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97834
--- Comment #4 from Roland Illig <roland.illig at gmx dot de> ---
(In reply to Nathan Sidwell from comment #3)
> gcov has its own buffer.
I wish the above were true. Using gcov from GCC 5.5.0, ktrace shows that the
actual buffer size is 0. That's consistent with the setbuf call in gcov_open.
When the instrumented binary is run, this is what happens:
8998 1 make CALL open(0x7affea387b00,0x202,0x1b6)
8998 1 make NAMI "/home/rillig/proj/src/usr.bin/make/arch.gcda"
8998 1 make RET open 3
8998 1 make CALL fcntl(3,F_SETLKW,0x7f7fffc0b6c0)
8998 1 make RET fcntl 0
8998 1 make CALL fcntl(3,F_GETFL,0)
8998 1 make RET fcntl 2
8998 1 make CALL __fstat50(3,0x7f7fffc0b6e0)
8998 1 make RET __fstat50 0
8998 1 make CALL read(3,0x7affea752497,1)
8998 1 make GIO fd 3 read 1 bytes
"a"
8998 1 make RET read 1
8998 1 make CALL read(3,0x7affea752497,1)
8998 1 make GIO fd 3 read 1 bytes
"d"
8998 1 make RET read 1
8998 1 make CALL read(3,0x7affea752497,1)
8998 1 make GIO fd 3 read 1 bytes
"c"
8998 1 make RET read 1
8998 1 make CALL read(3,0x7affea752497,1)
8998 1 make GIO fd 3 read 1 bytes
"g"
8998 1 make RET read 1
8998 1 make CALL read(3,0x7affea752497,1)
8998 1 make GIO fd 3 read 1 bytes
"*"
8998 1 make RET read 1
8998 1 make CALL read(3,0x7affea752497,1)
8998 1 make GIO fd 3 read 1 bytes
"5"
8998 1 make RET read 1
8998 1 make CALL read(3,0x7affea752497,1)
8998 1 make GIO fd 3 read 1 bytes
"0"
8998 1 make RET read 1
8998 1 make CALL read(3,0x7affea752497,1)
8998 1 make GIO fd 3 read 1 bytes
"5"
8998 1 make RET read 1
8998 1 make CALL read(3,0x7affea752497,1)
8998 1 make GIO fd 3 read 1 bytes
"\M-4"
8998 1 make RET read 1
8998 1 make CALL read(3,0x7affea752497,1)
8998 1 make GIO fd 3 read 1 bytes
"\M--"
8998 1 make RET read 1
8998 1 make CALL read(3,0x7affea752497,1)
8998 1 make GIO fd 3 read 1 bytes
"\^_"
8998 1 make RET read 1
8998 1 make CALL read(3,0x7affea752497,1)
8998 1 make GIO fd 3 read 1 bytes
"\M-N"
These syscalls obviously take more time than any form of buffering.
