Date: Tue, 29 Nov 2005 03:07:30 -0800 From: Maxim Sobolev <sobomax@portaone.com> To: phk@freebsd.org Cc: "current@freebsd.org" <current@freebsd.org> Subject: Correct usage of g_read_data() API Message-ID: <438C3672.1010700@portaone.com>
next in thread | raw e-mail | index | archive | help
Hi Poul, After making a fix for geom_label (rev.1.4 of g_label_ntfs.c) I have decided to check if there are any other places which don't check return values of g_read_data(9). What I found however is that there are several different usage patterns, some of which rely on undocumented internal properties of the said API, particularly: --- 1 --- error = random(); data = g_read_data(..., &error); if (error != 0) handle_error(error); --------- --- 2 --- error = random(); data = g_read_data(..., &error); if (buf == NULL || error != 0) handle_error(error); --------- --- 3 --- error = random(); data = g_read_data(..., &error); if (buf == NULL) handle_error(error); --------- According to my reading of g_read_data(9) only (3) is valid, since both (1) and (2) assume that error will always be cleared on successful completion, while (1) also assumes that error will be set to non-zero value in all cases when error happens and (2) assumes that non-NULL value can be returned even on error. Therefore, either documentation should be extended to match implementation or usage of g_read_data() fixed in all places where it is incorrect. Please advise. -Maxim P.S. I can do the actual changes by myself, both to documentation and/or to code.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?438C3672.1010700>