Hello Konstantin and Bruce, thank you for your comments. Clearly, this should have been addressed in the review and checked by a second pair of eyes. I mostly did the man page changes and can't comment much on the actual coding example. Two people (dab and jilles) approved the revision in the Phabricator review. There was plenty of time to review it between April 16 and May 15 when then the original submitter asked if there was anything else that needs to be done.
Now, having said that and the the change is now both in HEAD and stable/12, how do we proceed? Revert them both or patch those issues as follow-ups? Note that a src committer should do that and not me trying to add more EXAMPLE sections to man pages with my doc commit bit. Regards, Benedict Am 18.05.19 um 07:45 schrieb Bruce Evans: > On Sat, 18 May 2019, Konstantin Belousov wrote: > >> On Sat, May 18, 2019 at 03:15:08AM +0000, Benedict Reuschling wrote: >>> Author: bcr (doc committer) >>> Date: Sat May 18 03:15:07 2019 >>> New Revision: 347951 >>> URL: https://svnweb.freebsd.org/changeset/base/347951 >>> >>> Log: >>> MFC r347617: >>> Add small EXAMPLE section to bsearch.3. >>> >>> Submitted by: fernape (via Phabricator) >>> Reviewed by: bcr, jilles, dab >>> Approved by: bcr (man pages), jilles (src) >>> Differential Revision: https://reviews.freebsd.org/D19902 >>> >>> Modified: >>> stable/12/lib/libc/stdlib/bsearch.3 >>> Directory Properties: >>> stable/12/ (props changed) >>> >>> Modified: stable/12/lib/libc/stdlib/bsearch.3 >>> ============================================================================== >>> >>> --- stable/12/lib/libc/stdlib/bsearch.3 Sat May 18 02:02:14 >>> 2019 (r347950) >>> +++ stable/12/lib/libc/stdlib/bsearch.3 Sat May 18 03:15:07 >>> 2019 (r347951) >>> @@ -32,7 +32,7 @@ >>> .\" @(#)bsearch.3 8.3 (Berkeley) 4/19/94 >>> .\" $FreeBSD$ >>> .\" >>> -.Dd February 22, 2013 >>> +.Dd May 15, 2019 >>> .Dt BSEARCH 3 >>> .Os >>> .Sh NAME >>> @@ -83,6 +83,61 @@ The >>> function returns a pointer to a matching member of the array, or a null >>> pointer if no match is found. >>> If two members compare as equal, which member is matched is >>> unspecified. >>> +.Sh EXAMPLES >>> +A sample program that searches people by age in a sorted array: >>> +.Bd -literal >>> +#include <assert.h> >>> +#include <stdint.h> >>> +#include <stdio.h> >>> +#include <stdlib.h> >>> +#include <string.h> >>> + >>> +struct person { >>> + char name[5]; >>> + int age; >>> +}; > > This example has a high density of style bugs. kib pointed out some. > Some of the others are: > > (1) Not sorting the struct members on alignment. (style(9) says to sort > on use, then size, but means alignment). > (2) Not indenting the struct member names. > (3) Not use a prefix for struct member names. This is not important for > small programs. > >>> + >>> +int >>> +compare(const void *key, const void *array_member) >>> +{ >>> + int age = (intptr_t) key; >>> + struct person person = *(struct person *) array_member; >> These two lines contain at least three style(9) bugs, and at least one >> warning at higher warning level. >> >>> + >>> + return (age - person.age); >>> +} >>> + >>> +int >>> +main() >> Why use K&R definition ? >> >>> +{ >>> + struct person *friend; >>> + >>> + /* Sorted array */ >>> + struct person friends[6] = { >>> + { "paul", 22 }, >>> + { "anne", 25 }, >>> + { "fred", 25 }, >>> + { "mary", 27 }, >>> + { "mark", 35 }, >>> + { "bill", 50 } >>> + }; >>> + >>> + size_t array_size = sizeof(friends) / sizeof(struct person); >> Since you used const elsewere, why did not you used it there ? > > (4) The size expression is obfuscated by spelling the element size as > sizeof(struct person) instead of sizeof(friends[0]). > (5?) FreeBSD now spells this size expression as nitems(friends). I don't > like the unportability of this, especially in an example. > >> >>> + >>> + friend = bsearch((void *)22, &friends, array_size, sizeof(struct >>> person), compare); >> Taking address of an array is weird. >> Line is too long. > > (6) C99 specifies that the search is for a member of the array that matches > the object pointed to by 'key'. Here the key of (void *)22 is a > not a pointer to an object. It is a cookie which points to garbage. > The cookie is unique enough to work in practice. Perhaps you can > prove it to always work if the option type intptr_t is supported. > But this is a bad example. > (4a) same obfuscation of the element size. > >> >>> + assert(strcmp(friend->name, "paul") == 0); >>> + printf("name: %s\enage: %d\en", friend->name, friend->age); >>> + > > (7) Extra blank line. This is not too bad, but is not done consistently. > >>> + friend = bsearch((void *)25, &friends, array_size, sizeof(struct >>> person), compare); > > (8) More too-long lines. > (4b) More sizeof(person)'s. > >>> + assert(strcmp(friend->name, "fred") == 0 || strcmp(friend->name, >>> "anne") == 0); > > (7a) No extra blank line for "fred". > >>> + printf("name: %s\enage: %d\en", friend->name, friend->age); >>> + >>> + friend = bsearch((void *)30, &friends, array_size, sizeof(struct >>> person), compare); >>> + assert(friend == NULL); >>> + printf("friend aged 30 not found\en"); > > I didn't notice before that the key cookie was an encoding of the age. > > The key is not unique (there are 2 25's). This gives an example of low > quality data and the example has minimal error handling handling for this > without describing what it is doing (it asserts uniqueness for age 22 and > it asserts one of the 2 possibilities for age 25. It can only do this > because it knows too much about the data). > > It is more than a style bug to handle errors in data by assert() or > otherwise killing the program, except possibly when the data is supposed > to be good. > >>> + >>> + return (EXIT_SUCCESS); >>> +} >>> +.Ed >>> .Sh SEE ALSO >>> .Xr db 3 , >>> .Xr lsearch 3 , > > Bruce
signature.asc
Description: OpenPGP digital signature
