Add foreground server option

[andy-bower]

New '-F' option is like '-D' but does not fork. This enables improved init system integration methods and can make up for the lack of a PID file writing option (fixes: #1).

[solardiz]

What option letters do other services use for this? We may try to be consistent. (I put some related context and thoughts in #1.)

[solardiz]

What compilers recognize /* [[fallthrough]]; */?

FWIW, in LKRG we define and use this macro:

#if defined(fallthrough)
 #define P_FALL_THROUGH fallthrough
#elif defined(__GNUC__) && __GNUC__ > 6
 #define P_FALL_THROUGH __attribute__ ((fallthrough))
#else
 #define P_FALL_THROUGH /* FALLTHROUGH */
#endif

but its #if defined(fallthrough) part is specific to the kernel. Of course, I'd rather have just a simple comment understood by compilers than introduce a macro into this project as well... or avoid the fall-through to side-step it.

In yescrypt, including its integration in libxcrypt, it's just /* FALLTHRU */, and I don't recall warning reports about that, so apparently it works pretty well.

[andy-bower]

Heh, I initially put just [[fallthrough]] because the target compile had C23 and I thought "let's do it because I can" but then turned it into a comment for the PR here for compatibility - will take your advice!

[solardiz]

I wonder if we can reasonably minimize the code differences vs. LKRG logger.c here.

[andy-bower]

I didn't get why the first (unconditional) of the two setsid() calls was there in both tools. So it's possible I was wrong to make them conditional on not being in foreground mode.

[andy-bower]

What option letters do other services use for this? We may try to be consistent. (I put some related context and thoughts in #1.)

This is a good question. I agree -D is very common. (gpm, mini-httpd).

Quite a few daemons get used with -d as a debug mode that happens to achieve this effect. (unbound, lldpd).

in.tftpd offers -L for "listen". (This is not a bad choice.)

isc-dhcp-server uses -f. (I avoided this because of typical use for config files.)

havegd uses -F.

[andy-bower]

I force-pushed a revision to my PR simply to knock off the 3 blatant errors you picked up!

[solardiz]

This is a good question. I agree -D is very common. (gpm, mini-httpd).

Yes, and also sshd. If we started from scratch now, popa3d would probably not default to inetd mode since usage of (x)inetd became rather uncommon lately. So we could also have -i for inetd mode, like sshd has. And the default (no options) could be what we had as popa3d -D so far.

... except what to do if standalone mode support is not compiled in? Default to inetd mode in that case, or refuse to run unless -i is specified maybe? Edit: or no longer support compiling without standalone mode?

Given our past use of -D, how do we recover from this now? Go with your approach of adding an extra option such as -F for foreground mode or redefine the meaning of -D and document this prominently, e.g. along with at least numbering the next release at least 1.1 or even 2.0? (I'd also merge the Maildir support changes if so.)

[solardiz]

@vt-alt @ldv-alt What's your opinion on this, if any? Note that I already redefined -D in code reuse for LKRG logger for it to be consistent with sshd.

[andy-bower]

Hi,

This is a good question. I agree -D is very common. (gpm, mini-httpd).

Yes, and also sshd. If we started from scratch now, popa3d would probably not default to inetd mode since usage of (x)inetd became rather uncommon lately. So we could also have -i for inetd mode, like sshd has. And the default (no options) could be what we had as popa3d -D so far.

... except what to do if standalone mode support is not compiled in? Default to inetd mode in that case, or refuse to run unless -i is specified maybe?

Given our past use of -D, how do we recover from this now? Go with your approach of adding an extra option such as -F for foreground mode or redefine the meaning of -D and document this prominently, e.g. along with at least numbering the next release at least 1.1 or even 2.0? (I'd also merge the Maildir support changes if so.)

I suspect it would be a bad plan to invert the meaning of -D or change the default behaviour.

But that doesn't mean you have to go with my -F suggestion as you might have a nicer one in mind! (Other than -D! :-( )

And you could harmlessly add in new preferred options to explicitly select inetd mode and daemon mode and deprecate the old ways.

[solardiz]

I suspect it would be a bad plan to invert the meaning of -D or change the default behaviour.

Maybe this should depend on whether we're giving popa3d some maybe-final life support touches (conservative), or giving it a new life (modernized). Does a new life for a POP3-only server make sense these days (if we, say, added builtin TLS support)? Maybe not, and I see no reason to compete with Dovecot by adding IMAP. Also I don't currently have a use for popa3d myself, so am not motivated and would not dogfood it.

But that doesn't mean you have to go with my -F suggestion as you might have a nicer one in mind!

I currently don't have a better idea, and you found one precedent of -F.

[andy-bower]

I suspect it would be a bad plan to invert the meaning of -D or change the default behaviour.

Maybe this should depend on whether we're giving popa3d some maybe-final life support touches (conservative),

This sounds like the realistic perspective to me! 😁

or giving it a new life (modernized). Does a new life for a POP3-only server make sense these days (if we, say, added builtin TLS support)? Maybe not, and I see no reason to compete with Dovecot by adding IMAP. Also I don't currently have a use for popa3d myself, so am not motivated and would not dogfood it.

Might be worth ensuring this builds with C23/gcc-15 as part of the drop.