On Mon, Oct 26, 2020 at 02:05:28AM -0600, Scott Long wrote: > > > On Oct 26, 2020, at 1:50 AM, Baptiste Daroussin <b...@freebsd.org> wrote: > > > > On Mon, Oct 26, 2020 at 12:11:56AM -0600, Warner Losh wrote: > >> On Mon, Oct 26, 2020 at 12:01 AM Alex Kozlov <a...@freebsd.org> wrote: > >> > >>> On Sun, Oct 25, 2020 at 11:37:34AM +0100, Stefan Esser wrote: > >>>> Am 25.10.20 um 06:56 schrieb Alex Kozlov: > >>>>> On Sat, Oct 24, 2020 at 04:37:45PM +0200, Stefan Esser wrote: > >>>>>> Am 24.10.20 um 09:48 schrieb Alex Kozlov: > >>>> [...] > >>>>>>> You are hardcoding assumption that LOCALBASE = /usr/local. Please > >>> make it > >>>>>>> overridable with LOCALBASE environment variable. > >>>>>> This was a trivial change to get us going with calendars provided by > >>>>>> a port (which has not been committed, yet - therefore there are no > >>>>>> port-provided calendars, neither under /usr/local nor under any other > >>>>>> PREFIX, as of now). > >>>>> > >>>>>> I understand what you are asking for, but in such a case I'd rather > >>>>>> think you want to rebuild FreeBSD with _PATH_LOCALBASE modified in > >>>>>> paths.h. > >>>>> The PREFIX != LOCALBASE and both != /usr/local configurations > >>>>> are supported in the ports tree and the base for a long time, please > >>> see > >>>>> > >>> https://www.freebsd.org/doc/en_US.ISO8859-1/books/porters-handbook/porting-prefix.html > >>>> > >>>> Yes, and I do not need to look that up in the handbook, having been > >>>> a ports committer for 2 decades by now. > >>>> > >>>>> If after this commit you need to rebuild base to use non-default > >>> LOCALBASE/PREFIX > >>>>> it is pretty big regression and POLA. > >>>> > >>>> How is that any different than before? > >>>> > >>>> What I did is make the PATH easier to change when you rebuild base. > >>>> > >>>> There are numerous programs in base that contain the literal string > >>>> /usr/local - and what I did was implement a mechanism that allows > >>>> to replace this literal reference with a simple change in paths.h. > >>>> > >>>> If you do not modify paths.h for a different LOCALBASE, then you'll > >>>> get a wrong _PATH_DEFPATH compiled into your binaries, for example. > >>>> > >>>>>> And I have made this a single instance that needs to be changed. > >>>>>> Before my change there were 2 instances of /usr/local hard-coded > >>>>>> in _PATH_DEFPATH - now you have to only change the definition of > >>>>>> _PATH_LOCALBASE to adjust all 3 locations that use it. > >>>>> I think you made situation worse, there were two stray hardcoded > >>>>> string and now there is official LOCALBASE define which likely will be > >>>>> used by other people in the future. > >>>> > >>>> I'd hope so to get rid of many of the 1713 literal uses of /usr/local > >>>> in our source tree. > >>>> > >>>>>> If you can show me precedence of a LOCALBASE environment variable > >>>>>> being used in the way you suggest, I'd be willing to make calendar > >>>>>> use it. > >>>>> Just an analogy from LOCALBASE make variable, perhaps CALENDAR_HOME > >>>>> is a better name. > >>>> > >>>> Yes, I already suggested CALENDAR_HOME, but as an environment variable > >>>> to check, if you want to be able to path an additional directory (or > >>>> search path) to the calendar program at run-time. But why introduce > >>>> a CALENDAR_HOME macro in the sources, if the port supplied calendar > >>>> files are known to be found at LOCALBASE/share/calendar (for some value > >>>> of LOCALBASE). > >>>> > >>>> I want to make more programs that currently hard-code /usr/local use > >>>> _PATH_LOCALBASE instead. This C macro can then be default to /usr/local > >>>> but can be overridden by passing LOCALBASE to the compiler (from the > >>>> build infrastructure) when paths.h is included. > >>>> > >>>> Instead of referring to _PATH_LOCALBASE these files could directly use > >>>> LOCALBASE, but since other paths are defined as _PATH_xxx in paths.h I > >>>> think it is best to follow this precedent. > >>>> > >>>>>> But then I think a CALENDAR_HOME variable would be even more useful, > >>>>>> since it would allow to search an additional user selected directory > >>>>>> (and not just share/calendar within what you provide as LOCALBASE). > >>>> > >>>> My change did not add any dependency on LOCALBASE to any previously > >>>> existing functionality. It added support for calendar files provided > >>>> by a port (a feature that did not exist before) at a location that is > >>>> correct for the big majority of users (who do not modify LOCALBASE). > >>>> > >>>> As I said: I'm going to make it easier to build the base system with > >>>> a different LOCALBASE, but not by run-time checking an environment > >>>> variable that specifies LOCALBASE in each affected program. > >>> It seems that you intend to follow through no matter what. So, just for > >>> the record, I think that hardcoding LOCALBASE and requiring base rebuild > >>> to change it is a very wrong approach. > >>> > >> > >> So, first off, it's already hard coded. Stefan's changes change the hard > >> coding from 'impossible to change' to 'changeable with a recompile' which > >> is an improvement. It might even wind up as a build variable (or not, doing > >> that has some really ugly, nasty dependencies). > >> > >> But even in ports-land, it's a compile time constant. Quite a large number > >> of ports will allow you to change it at compile / build time, but not > >> after. You have to rebuild if you want to change PREFIX... > >> > >> So I'm a bit puzzled what makes this the wrong approach? > >> > > > > I think what Alex revents to is the following: > > > > Some utilities in base base either have a configurable way to look for > > things in > > localbase (via configuration entries for instances): > > - syslog > > - periodic > > - rc > > - man > > Some have hardcoded LOCALBASE but only after looking first at the LOCALBASE > > env > > var: > > - usr.sbin/pkg > > - mailwrapper > > > > which means with a prebuilt base I can still rebuild all my packages with a > > different localbase and it will work with only a few configurations changes. > > which imho is a good target. > > > > The list of tools which hardcodes /usr/local > > - calendar > > - fortune > > - cron > > - bsnmp > > - nvmecontrol > > - cpucontrol (at least can be workaround via -d option) > > > > > > It would be pretty trivial to add a new libc function, something like > getlocaldir.2, > that took care of searching the environment and the invoking a fallback to the > compile-time default from path.h. I’ll see if I can come up with something > for > review before I fall asleep.
Exactly what I was thinking about ;) could also be a simple static inline function somewhere (path.h?) if we don't want to "pollute" libc. I am fine with both. Best regards, Bapt
signature.asc
Description: PGP signature