These defines are used both in usr/src/head/Makefile and usr/src/man/man1m/Makefile. Do we have any other Makefile, ? Of course, I could rename defines, so that they have less general names, for example, prefix them with "USERDEFS_*". Or should I put these defines in some new Makefile and include it in these two?

I don't understand. sysconfig() returns values of predefined constants from other files. Why do we need one more indirection level? These are really constants, just different distributions want to set them to different values.

Or do you suggest the following? If we create new function and then use #define  DEFLOCK_AFTER_RETRIES userdefs(DEFLOCK_AFTER_RETRIES) in userdefs.h, we can use compile-time variables (-DDEFLOCK_AFTER_RETRIES=$(DEFLOCK_AFTER_RETRIES)) while compiling source code of this function to avoid using sed? But what about man files?

Well it's quite possible I'm the one who is confused :)  so let me ask:

Are the values provided by these userdefs.h macros really constants?

Or are these macros the mechanism by which consumer programs get the

"current setting for homedir" etc?

And a related question:  How should this work?

Perhaps DEFPARENT should be the currently configured default (from /etc/default/something?)

and then userdefs.h might have something like:
extern const char *_userdefs(const char *);
#define DEFPARENT   (_userdefs("DEFPARENT"))

Or maybe the _userdefs() arg is an int.  whatever.

These values, provided by userdefs.h now, are constants. What we do is just introducing a easy way to redefine them at compile time (like we do for PYTHON_VERSION). Initial idea was to allow redefining base_dir at runtime - http://buildzone.oi-build.r61.net/webrev-7648/ . But Peter saw inconsistency in fact that they could be redefined in several places (/usr/sadm/defadduser and /etc/default/useradd).

I guess some of what is troubling me is that the issue has no details about:

What's there now and how that was meant to work; what's wrong with what's there now, and how the proposed fix makes that better.

I've updated the issue. Can we have the first part of the issue (compile time customization) fixed? If so, how this particular patchset should be updated? I consider that run-time customization is good, but can be a separate issue. I'm not especially interested in fixing second part.  What I want is to make 'useradd -m test' work out of the box without impact on other distributions and without hacky sollutions. I prefer if it continued working with self-compiled illumos-gate. As I see, DEF constants, defined in userdefs.h, are used only in user-utilities.
If we proceed with this review... 1) Should I rename constants in Makefile.master so that they use less generic names? 2) Should I move these definitions outside Makefile.master? If yes, then where? 3) Should I remove this header from files where it's not used?

I realize you have probably lost interest in this issue, but I finally put up an alternative.

See:  https://www.illumos.org/rb/r/420/