On Nov 23, 2017, at 2:12 PM, Richard Sharpe <realrichardsharpe@xxxxxxxxx> wrote:
> I am running into problems with this in my latest build:
>
> #if defined(_WIN32)
> char * cap_pipe_buf; /**< Pointer
> to the buffer we read into */
> DWORD cap_pipe_bytes_to_read; /**< Used by
> pipe_dispatch */
> DWORD cap_pipe_bytes_read; /**< Used by
> pipe_dispatch */
On Windows, we read from a pipe with ReadFile(), the byte count argument to which is a DWORD, and the "bytes I actually read" argument to which is a pointer to a DWORD.
> #else
> size_t cap_pipe_bytes_to_read; /**< Used by
> pipe_dispatch */
> size_t cap_pipe_bytes_read; /**< Used by
> pipe_dispatch */
On UN*X, we read from a pipe with cap_pipe_read(), which calls ws_read(), which is just a #define for read on UN*X; the byte count argument to read() is a size_t, and the "bytes I actually read, or -1 for error" return value for which is an ssize_t.
What are the problems that this causes?
Perhaps we should simply restrict the number of bytes to be read in any one call to INT_MAX, which should fit in a DWORD, a size_t, or an ssize_t, and thus work on Windows and UN*X, either 32-bit or 64-bit, and declare both cap_pipe_bytes_to_read and cap_pipe_bytes_read to be int.