A request for opinion on proc read xlx integer in van Vliet

[Jose E. Marchesi]

> Hello again;
>
> This appears in van Vliet's transput:
>
>           #<#{PROC read xlx integer = ({REF #L# {INT i) {BOOL:
>             # if an integer is read successfully,
>               it is assigned to 'i', and the routine returns true;
>               otherwise, it returns false #
>             {BEGIN^BOOL ok:= {TRUE, {BOOL neg = k = "-";
>               (neg {OR k = "+" | skip spaces);
>               ({NODIGIT k | mend char(radix10digit, "0"));
>               #L# {INT j:= #K# chardig(k);
>
>               {WHILE
>                 {IF status {OF cover {OF f {SAYS line ok
>                 {THEN get char(f, k);
>                   {IF^NODIGIT k
>                   {THEN backchar(f); {FALSE
>                   {ELSE (ok | ok:= j {ADD chardig(k)); {TRUE
>                   {FI
>                 {ELSE^FALSE
>                 {FI
>               {DO^SKIP^OD;
>               {IF ok {THEN i:= (neg | -j | j) {FI;
>               ok
>             {END#>#;
>
> This will expand to
>
> proc read_short_short_integer(ref short short int i) bool: ...
> proc read_short_integer(ref short int i) bool: ...
> proc read_integer(ref int i) bool: ...
> proc read_long_integer(ref long int i) bool: ...
> proc read_long_long_integer(ref long long int i) bool: ...
>
> This "proc" is immediately followed by
>
>           {PROC read l real = ({UNION(#<#{REF #L# {REAL#>#) r) {BOOL:
>             # similar to 'read xlx integer', for real numbers #
>
> I'm not going to copy/paste the rest of it.  Note that it chooses to take a
> union of the various lengths of reals rather than creating
>
> proc read_real(ref real r) bool:...
> proc read_long_real(ref long real r) bool: ...
> proc read_long_long_real(ref long long real r) bool: ...
>
> I find this kind of awkward.  He only references read_xlx_integer once, in
> the body of read_l_real.  But even so.
>
> I would appreciate any opinions offered!

I don't know why van Vliet decided to use an union in this case, rather
than expansion.  I think we should prefer expansion for efficiency
reasons: balancing is done at compile-time.