[Rd] Use d0 instead of _8 in ppr.f ?

[Martin Maechler]

I had chosen _8  consciously,  after reading a bit up at
constants in Fortran (in the year 2026, more than 50 years after
the 77 of 'Fortran 77') ... notably after
getting Fortran compiler warnings from flang 22.

I do think once you get very slightly used to this, using *_8 is
*faster* to parse for a human than d0  (to which many of course
has been getting used to ...).

I'm a bit sorry that this really seems to "need" an R-devel post
..., really?

Martin Maechler

>>>>> Suharto Anggono Suharto Anggono via R-devel 
>>>>>     on Wed, 29 Apr 2026 09:59:40 +0000 (UTC) writes:

    > SVN revision 89917 adds "_8" to some numbers in src/library/stats/src/ppr.f , with message
    > flang-22 wants explicit "double" / "8-bit float" constants

    > How about using "d0" instead? It is more consistent within the file and a little more portable. The number doesn't have to be equal to the number of bytes, 8.

    > Here is a patch that replaces "_8" with "d0".

    > --- ppr.f	2026-04-29 09:39:32.145702763 +0000
    > +++ ppr_d0.f	2026-04-29 09:39:32.268715531 +0000
    > @@ -842,9 +842,9 @@
 
    > data df, gcvpen, ismethod, trace /4d0, 1d0, 0, .false./
    > data ifl,maxit, conv, mitone, cutmin, fdel
    > -     &    / 6,  20,  .005_8,   20,   0.1_8, 0.02_8 /
    > +     &    / 6,  20,  .005d0,   20,   0.1d0, 0.02d0 /
    > data span,alpha, big,   cjeps,  mitcj, lf
    > -     &    / 0.0, 0.0, 1.0e20, 0.001_8,   1,   2 /
    > +     &    / 0.0, 0.0, 1.0e20, 0.001d0,   1,   2 /
    > end
 
    > subroutine setppr(span1, alpha1, optlevel, ism, df1, gcvpen1)
    > @@ -1259,7 +1259,7 @@
    > common /spans/ spans /consts/ big,sml,eps
 
    > data spans,               big,  sml,  eps
    > -     &  /0.05_8, 0.2_8, 0.5, 1.0e20,1.0e-7,1.0e-3/
    > +     &  /0.05d0, 0.2d0, 0.5, 1.0e20,1.0e-7,1.0e-3/
    > end
    > c---------------------------------------------------------------
    > c
    > @@ -1384,7 +1384,7 @@
    > c     tol for 'spar' estimation:
    > param(3) = 1d-2
    > c     'eps' (~= 2^-12 = sqrt(2^-24) ?= sqrt(machine eps)) in ./sbart.c :
    > -      param(4) = .000244_8
    > +      param(4) = .000244d0
 
    > ier = 1
    > call rbart(gcvpen,df1,dx,dy,dw,0.0d0,n,knot,nk,coef,dsmo,lev,


    > This is an alternative that uses "d" for all "double precision".

    > --- ppr.f	2026-04-29 09:39:32.145702763 +0000
    > +++ ppr_d0_all.f	2026-04-29 09:39:32.291717918 +0000
    > @@ -842,9 +842,9 @@
 
    > data df, gcvpen, ismethod, trace /4d0, 1d0, 0, .false./
    > data ifl,maxit, conv, mitone, cutmin, fdel
    > -     &    / 6,  20,  .005_8,   20,   0.1_8, 0.02_8 /
    > +     &    / 6,  20,  .005d0,   20,   0.1d0, 0.02d0 /
    > data span,alpha, big,   cjeps,  mitcj, lf
    > -     &    / 0.0, 0.0, 1.0e20, 0.001_8,   1,   2 /
    > +     &    / 0d0, 0d0, 1.0d20, 0.001d0,   1,   2 /
    > end
 
    > subroutine setppr(span1, alpha1, optlevel, ism, df1, gcvpen1)
    > @@ -1258,8 +1258,8 @@
    > double precision spans(3),    big,sml,eps
    > common /spans/ spans /consts/ big,sml,eps
 
    > -      data spans,               big,  sml,  eps
    > -     &  /0.05_8, 0.2_8, 0.5, 1.0e20,1.0e-7,1.0e-3/
    > +      data spans,                 big,  sml,  eps
    > +     &  /0.05d0, 0.2d0, 0.5d0, 1.0d20,1.0d-7,1.0d-3/
    > end
    > c---------------------------------------------------------------
    > c
    > @@ -1384,7 +1384,7 @@
    > c     tol for 'spar' estimation:
    > param(3) = 1d-2
    > c     'eps' (~= 2^-12 = sqrt(2^-24) ?= sqrt(machine eps)) in ./sbart.c :
    > -      param(4) = .000244_8
    > +      param(4) = .000244d0
 
    > ier = 1
    > call rbart(gcvpen,df1,dx,dy,dw,0.0d0,n,knot,nk,coef,dsmo,lev,

    > ______________________________________________
    > R-devel using r-project.org mailing list
    > https://stat.ethz.ch/mailman/listinfo/r-devel