From: Wilson Bilkovich <wilsonb@...>
Date: 2006-04-25T12:31:25+09:00
Subject: Re: Format of RUBYOPT

On 4/24/06, ara.t.howard@noaa.gov <ara.t.howard@noaa.gov> wrote:
> On Tue, 25 Apr 2006, Ryan Leavengood wrote:
>
> > On 4/24/06, ara.t.howard@noaa.gov <ara.t.howard@noaa.gov> wrote:
> >>
> >> if you look in ruby.c you'll see that it's literally parsed as part of the
> >> commandline - except only the switchs 'IdvwrK' and 'T' are accepted.  it looks
> >> like it is indeed space separated.
> >>
> >>    harp:~ > cat rubylib/a.rb
> >>    A=42
> >>
> >>    harp:~ > RUBYOPT="-ryaml -Irubylib/ -ra" ruby -e "p [YAML, A]"
> >>    [YAML, 42]
> >>
> >> seems to work.
> >
> > Hmmm, OK. Then why does setting RUBYOPT to just rubygems work (it has
> > no dash?) That doesn't work on the command-line (at least on Windows):
> >
> > C:\_Ryan\ruby_installer\svn\NSIS>ruby rubygems -e 'puts "hello"'
> > ruby: No such file or directory -- rubygems (LoadError)
> >
> > C:\_Ryan\ruby_installer\svn\NSIS>ruby -rubygems -e 'puts "hello"'
> > hello
> >
> > Maybe to be safe I should always just add -rubygems (i.e with a dash prefix.)
>
> ruby.c, in proc_options:
>
>
>      731     if (rb_safe_level() == 0 && (s = getenv("RUBYOPT"))) {
>      732         while (ISSPACE(*s)) s++;
>      733         if (*s == 'T' || (*s == '-' && *(s+1) == 'T')) {
>      734             int numlen;
>      735             int v = 1;
>      736
>      737             if (*s != 'T') ++s;
>      738             if (*++s) {
>      739                 v = scan_oct(s, 2, &numlen);
>      740                 if (numlen == 0) v = 1;
>      741             }
>      742             rb_set_safe_level(v);
>      743         }
>
>
> in the else block we eat the dash and any space, iff present.  then we look
> for the opt char
>
>
>      744         else {
>      745             while (s && *s) {
>      746                 if (*s == '-') {
>      747                     s++;
>      748                     if (ISSPACE(*s)) {
>      749                         do {s++;} while (ISSPACE(*s));
>      750                         continue;
>      751                     }
>      752                 }
>      753                 if (!*s) break;
>      754                 if (!strchr("IdvwrK", *s))
>      755                     rb_raise(rb_eRuntimeError, "illegal switch in RUBYOPT: -%c", *s);
>
>
> now, given RUBYOPT=-rubygems, and *s==r, we call moreswitches
>
>
>      756                 s = moreswitches(s);
>      757             }
>      758         }
>      759     }
>
>
> so in this function
>
>
>      433 static char*
>      434 moreswitches(s)
>      435     char *s;
>      436 {
>      437     int argc; char *argv[3];
>      438     char *p = s;
>      439
>      440     argc = 2; argv[0] = argv[2] = 0;
>      441     while (*s && !ISSPACE(*s))
>      442         s++;
>      443     argv[1] = ALLOCA_N(char, s-p+2);
>      444     argv[1][0] = '-';
>      445     strncpy(argv[1]+1, p, s-p);
>      446     argv[1][s-p+1] = '\0';
>      447     proc_options(argc, argv);
>      448     while (*s && ISSPACE(*s))
>      449         s++;
>      450     return s;
>      451 }
>
>
> we eat whitespace, setup argv=[0, "-rubygems", 0] and argc == 2.  then recurse
> into proc_options (the first method shown).
>
>
> so, as far as i can tell (and i've had three beers), proc_options finds the
> 'r', jumps into moreswitches which restores the '-', and the were back into
> proc_options with '-rubygems', which works because of the 'ubygems' hack.  the
> parsing of RUBYOPT looks as if it ignores '-' or lack thereof be design, but i
> certainly can't be knowing.
>
> good luck!  ;-)
>
> -a
> --

To summarize, if you're going to alter RUBYOPT by appending something,
you should append a space and then '-rrubygems', to be sure?