On Sun, Jan 26, 2020 at 2:00 PM J William Piggott <elseifthen@xxxxxxx> wrote: > > > > On Sat, 25 Jan 2020, Karel Zak wrote: > > > On Fri, Jan 24, 2020 at 04:16:47PM -0300, Carlos Santos wrote: > > > >> That's a problem. It makes hwclock hard to include in embedded systems > >> due to the GPLv3 restrictions. > >> > >> I noticed that it comes due to sys-utils/hwclock-parse-date.y, which > >> was taken from gnulib. Would it be possible to take the file from an > >> previous version of gnulib that was still under GPLv2? > >> > >> An alternative approach would be porting a similar code using a more > >> liberal license, e.g. BSD. > >> > >> What do you think? > > > > I have tried to export it from gnulib with v2, but it was impossible > > by official gnulib tools. Maybe do it manually from some old > > tarball. I'll accept a patch for this if you have time do it. > > You do realize that I had to heavily modify that file to remove its > gnulib dependencies (because you said no to gnulib). If I recall > correctly I had newer and older versions to chose from and picked that > one due to it having the most bugs fixed while still being practical to > strip its gnulib dependence. > > The reasons for making the change were: > * remove hwclock's dependence on date(1) > * remove an insecure call to date(1) > * I thought there would be to many complaints if the accepted input > date formats were changed > > As to the last bullet point; personally I think having the --date option > accept every date syntax know to history is nonsense. A fixed format > would be fine with me. Since we switched hwclock's output to ISO 8601 > that seems like a good choice for its input. Or you could just use the > existing utillinux date parser. > > The question is, do you want to deal with any pushback for > changing the long established accepted --date formats? > > > > > I'll like to release 2.35.1 ASAP (due to bug in sfdisk --move-data), > > so we can add this license change too. Can't we use getdate(3), instead? I see some advantages in this approach: - It's a POSIX standard - It's supported by GLIBC, uClibc-ng and musl, among others - No need to maintain the code, since it is provided by the system libc - It's easy to a default template to use if DATEMSK is not defined (e.g. start with a copy of NetBSD's /usr/share/examples/getdate/datemsk.template and expand it as necessary). Being non-reentrant is irelevant for hwclock and the cost of reading an external file is negligible. -- Carlos Santos <unixmania@xxxxxxxxx>
