CHANGES

v45 changes (08/26/2026)

    Implemented commands for factorial, permutations and combinations.

    Minor wordsmithing and proofreading.



v44 changes (05/20/2026)

    Makefile:  removed the references to /usr/local for the mpdecimal
    library.  I've installed mpdecimal from the deb.sury.org PPA
    instead.

    The files in the debian subdirectory now cleanly build source
    and binary packages, which pass lintian checks.  Still haven't
    decided about how, or whether, to package for debian.  Currently
    it's set up as a "native" package.  That might change.

    Added missing Copyright and license declarations to Makefile, scripts,
    README, etc.

    Autoprinting can now print more than just the top of stack.  The
    "autoprint" command is no longer a boolean toggle:  the argument
    is the number of stack entries to be shown when autoprinting.

    Enabled -O2 for better diagnostics, and then added targeted warning
    suppression in the code via #pragma lines.

    Added missing #!  line to rca_cofloat, and changed rca_float from
    #!/bin/sh to #!/bin/bash due to use of <<< operator.

    The pop command no longer modifies the lastx value.  This lets
    "pop ly lx" act as undo.

    Corrected and improved man page and help for the new % commands.



v43 changes (04/09/2026)

    new commands:
      - added percent operator, and friends.
	percent is now '%', and modulo is now 'mod'.  required changes
	in parse_token() to support "+%" and "-%".  (that's gonna need
	more work at some point.)

	operators (all 2 operand) are:
	     %		r = y * x/100
	    +%		r = y + (y * x/100)
	    -%		r = y - (y * x/100)
	    %?		r = 100 * (x - y)/y  (the inverse of the previous two)

      - added "C", to select a currency floating point mode, and the
	corresponding display-as command "c".  "C" forces 2 decimals
	for both display, and for value storage on the stack.  the "2"
	comes from the locale, from frac_digits, so might be 0 or 3 in
	Japan or Kuwait, for example.

      - added 'm' (or "max") command, which prints as float, using
        max_digits

      - add "lasty"/"ly" commands, similar to "lastx"/"lx"

      - added built-in constants "inf" and "nan" (debug only)

    new behavior:
	"fixed" format now switches to "auto" when numbers get too big.

    revised the conversions needed for integer mode.  negative modulo
    values were causing mpd_get_u64() issues.

    various bug fixes

    man page:
	- moved some things from the intro into the RPN section
	- documented the half-up rounding policy
	- documented the "2 bit" trap with the "bits" command, and the
	  limit on entering numbers in non-decimal base format.
	- rearranged some info about integer mode
	- documented new commands

    cosmetic changes:
	major rearrangement of early function declarations, globals.
	added comments, reordered many routines more logically, and
	into categories.  rearranged the entries in the command table,
	with hopefully better headings.

    debug changes:
	- reintroduced debug-only raw stack dump in "state" command
	- added accounting to verify math on temp buffer sizing



v42 changes (04/03/2026)

    revised README.md, and clarified some of the build instructions
    in the Makefile.

    now dynamically size buffers used for printing, and calculate pi
    on startup, rather than initializing from a string constant, both
    allow rca to work for any number of digits.  also, make the Taylor
    iteration limit dependent on max_digits

    remove the cap on max_digits, as set by $RCA_DIGITS

    introduce a safe_snprintf() wrapper

    made "working" digit margins bigger, and tuned the atan algorithm
    cutoffs

    now use a context with more precision while initializing all of
    the constants (including pi)

    brought debian packaging metadata up to date

    added a state-of-the-branches summary, to both the mpdecimal
    and main branches.


v41 changes (04/01/2026)

    implemented trig functions with taylor series using the mpdecimal
    library.  cos and atan use the series, the rest use identity
    equations.  the conversion to mpdecimal is now complete.

    rationalized the use of DIGITS/max_digits for setting various
    limits.  there are now separate cutoffs for a) user visibility
    (i.e., rounding), b) some internal calculations (e.g., when to
    consider a taylor series complete), and c) the full precision that
    mpdecimal uses.  the 'a' level (30, by default) is "max_digits":
    it's what the user sees, and can be adjusted using $RCA_DIGITS.
    it controls display rounding, and also the precision used when
    doing equality comparisons.  the 'b' level (4 more, or 34, by
    default) is used by the trig functions, because they're so
    sensitive to tiny input values.  the 'c' level (8 more, or 38, by
    default) is what mpdecimal uses.  that feels like a lot of
    headroom, but it's cheap.

    added more protection against argument aliasing when using the
    mpdecimal API internally.  if you call foo(a, a, ctx), and foo is
    declared as foo(r, x, ctx), then if foo() modifies r, perhaps with
    an intermediate result, before finishing its calculations, it will
    change the value of x, which may still be needed.  protect by
    taking a local copy of x inside of foo(), and using that rather
    than the original x.

    valgrind now auto-enables (for tests) if it's present on the system

    adjusted precedence command output so all single char commands are
    first on line.

    adjusted defaults in rca_float/rca_cofloat

    proofread the man page again


v40 changes (03/27/2026)

    (intentional gap in the release numbers, in case I feel the need
    to do more changes to the floating point version.)

    *lots* of churn, in order to reimplement floating point using the
    mpdecimal library, eliminating issues caused by binary/decimal
    conversion with "normal" floating point.

    mpdecimal doesn't provide trig functions, so those are still
    implemented using native floating point.  as a result, the default
    number of significant digits in the calculator is configured as
    LDBL_DIG (18, usually), which is all that the trig functions can
    support.  $RCA_DIGITS can be used to increase this, at the risk of
    errors when working with trig results.

    mpdecimal is heavy on malloc/free, so there are now valgrind()
    checkpoints sprinkled in the code to help find the source of
    leaks.  all of our tests now run under valgrind, if it's present.


v25 changes (03/21/2026)

    reworked floating point tweaking to be fully based on display
    precision.  it feels a little weird, but i think it will help with
    the artifact problem.  they'll still be there -- just harder to
    see.  "48 47.8 - .2 - 0 ==" (aka "(((48 - 47.8) - .2) == 0)") didn't
    used to work.  the precision chose for rounding is halfway between
    your chosen precision, and the maximum available.

    added "bitc", a bitwise op that counts the 1 bits.

    reimplemented ';' as syntax, instead of as operator.  the elegance
    was nice, and the code is definitely more complicated, but this
    lets chained sub-expressions referencing lastx work:  (1 + 3; 7 +
    lx) == 11.  previously this gave an answer of 7 (because lx was
    defaulting to 0).  The RPN behavior of ';' is now less
    interesting.  it used to discard y, which i actually found useful
    once in a while.  now it just discards x, just like pop.

    removed declaration immediately following case-labels, added
    the -pedantic flag to catch this in the future

    cleaned up man page



v24 changes (03/15/2026)

    added new commands for degrees<->degrees.minutesseconds
    conversions:  "dd2dms" and "dms2dd"

    created new "Variadic" sections in help and man page

    added new standard deviation command:  "stddev"

    augmented stack printouts with the position of mark.  fixed bugs in
    mark handling, and improved user reporting for variadic
    (sum/avg/stddev/snapshot) commands.  added new "clearsnapshot"
    command

    major overhaul of "state" command output.  split some out for
    debug mode only, reduced verbiage, made friendlier.



v23 changes (03/10/2026)

    repurposed nop() from helping implement unary plus to actually
    doing nothing.  unary plus doesn't actually need nop().  and by
    turning "nop" into a silent pseudo-op, it can be used to silence
    output from a line of input.

    added new script, rca_cofloat.  like rca_float, but faster and
    more accurate.   also harder to use, and possibly more fragile.

    rca_float:  add caution about accuracy, bump default to 4 digits.
    fixed bug with ever-growing RCA_INIT value.

    add debian packaging support.  fixed install target to properly
    honor DESTDIR

    don't fail build due to copyright check, since there's a high
    chance the builder won't be able to fix it.

    fixed possible array/buffer overruns, found with more gcc warnings

    revised README.  add tools to help create a web-based README that
    looks like github's.  turns out the easiest way to get "github
    flavored markdown" is to use a github api and let them format it.
    then some extra sauce needs to be added to make it look pretty.



v22 changes   (03/05/2026)

    added "echo" toggle command, to explicitly control whether input
    from a file or pipe gets copied to the ouput.  it's more normal to
    not emit a copy of that input, but rca has always done so in the
    past, and the unit tests are written assuming it does.

    simplified input path, by skipping the readline/editline code
    entirely if input isn't coming from a terminal.  command line
    editing support code is now isolated in a single ifdef.

    added support for the editline command-line editing library.
    readline is GPL3, which means binaries linked against it become
    subject to the GPL3 distribution terms.  editline is the safer
    choice for a BSD-licensed program like rca.  the Makefile
    autodetects editline, and builds accordingly.  using readline is
    still easy, but requires specific user action (as described in
    Makefile comments).

    revised version handling somewhat


v21 changes   (03/02/2026)

    implemented a full-time infix mode.  invoke with "1 infix".  once
    in infix mode, all input is taken as infix format.  to enter
    anything else (for printing, configuration, quitting, etc, or
    RPN), prefix it/them on the line with ':'.  so ":0 infix" to leave
    the mode, ":q" to quit, etc.

    adding infix mode required a closer look at the shunting yard
    code, so many improvements there:  better accounting of
    parenthesis nesting levels (better error messages), and also code
    rearrangement so it more closely follows Dijkstra's original
    algorithm.  enabling the token and shunting categories for debug
    tracing ("rca 1 debug 5 tracing 1 infix") now shows a nice
    progression of the algorithm.

    UI change!  changed the "a", "z", "s", and "right" commands to
    "ap", "zf", "sep", and "ra".  this makes the printing and
    mode-change commands the only one-character commands:  f, d, u, o,
    h, b and F, D, O, H, B.

    added a new "clearvariables" command.

    wrote new code to implement numeric separators for decimal floats
    and integers, so no more dependence on the C library's %' feature.
    the former ifdefs, based on library type, seemed fragile.  the new
    defaults for whether separators are enabled are much cleaner:  we
    now trust and use what the locale gives us completely, unless it's
    the minimal C locale, in which case we augment it with ',' (for
    grouping) and '$'.

    made some adjustments to which commands do and don't cause
    autoprinting.

    made valgrind easier to enable for "make gentest"



v20 changes   (2/22/2026)

    cleaned and trimmed the "state" command enough for it to be a
    "real" command, not just debug.

    set max_digits and epsilon variables based on the constants in
    float.h, rather than hand-calculating.

    created separate reference files for the tests for each FP layout.
    there's now a simple ID, obtained from the "state" command, which
    lets us select which test files to use for each build type.

    added some useful compiler warnings to the build

    now much more careful with long long / ldouble conversions.  the
    RPi4 (much larger FP mantissa) and RPi3 (much smaller) are more
    sensitive to this than x86_64.  this involved more casting, but
    also more careful handling of transitions between long double,
    long long, and unsigned long long.  introduced new tests,
    especially for the bitwise operators.

    "bits" is now a (very natural) synonym command for "width"

v19 changes   (2/17/2026)
    major change to snapping/rounding tactics:  we now only tweak
    displayed floats, _never_ the stack.  this cleans up the remaining
    important testing differences between PC and ARM.  also, the tweak()
    routine is simpler, and better described.

    debugging commands are now hidden:  there are enough of debug
    commands that it's worth decluttering the help.

    added "epsilon" and "tweak" debug commands, to support somewhat
    architecture-independent tests

    Fixes for raw floating hex input and output.

    "-1 width" and "-1 digits" now result in maximum permitted value.

    debug tracing levels are now a bitmap, rather than a verbosity
    level.  selections for tracing are:  token flow, execution flow,
    the shunting yard algorithm, or any combination.


v18 changes   (2/12/2026)
    added rotate right, rotate left bitwise operators

    added "config" command, to summarize current settings

    fixed right alignment of binary format output

    eliminated trailing whitespace created by precedence,
    help, config, and license commands.  it was problematic for
    checking the output into git (for tests), and for possibly for
    the tests themselves.

    simplified code for suppressing early output from RCA_INIT


v17 changes   (2/10/2026)
    new "engineering" display format, new "digits" command, and
    renaming of existing display formats result in UI breakage:
    "precision", "decimals", "k", and "K" commands are gone, replaced
    by new no-argument commands:  "automatic/auto", "fixed", and
    "engineering/eng".  a new "digits" command specifies the number of
    floating point display digits.

    new snapshot/restore commands, for saving/restoring the entire stack

    rca_float: fixed handling of RCA_INIT, and rca's stderr

    improved rca's handing of input parsing errors.

    added Makefile install/uninstall targets (default to /usr/local)


v16 changes   (2/5/2026)
    removed unsigned math, and unsigned display mode.  neither make
    sense in the context of a practical calculator.

    improved help, and the way version numbers are maintained.


v14 change   (2/2/2026)
    implement option of right alignment for output values

    refactored how command output is managed.  commands are now
    silent unless they were entered last on a line.


v13 changes   (1/30/2026)
    octal input/output now use the "0o" prefix instead of just "0"

    added "zerofill", for left-filling in hex, binary, octal modes

    added unit conversion for mpg to l/100km

    some churn for deploying on github


v12 changes   (1/28/2026
    lots of work on precedence and associativity

    introduced the ";" operator to allow multiple expressions in (...)

    added dynamic variables (e.g., "_myvar"), eliminated r1-r5, s1-s5

    more rigorous about catching infix syntax errors

    better tracing support

    added "exp" operator

    "lastx" now works in infix


v10 changes   (1/13/2026)
    added "avg" command (similar to "sum")


v9 changes   (1/9/2026)
    "groups" command renamed to "separators"


v8 changes   (1/1/2026)
    no more hidden commands
    readline command completion
    tests/verification of poles/nan/inf for tan, powl


v7 changes   (12/30/25)
    standardize error messages
    add raw floating hex support
    improve "mark" command

.....


Copyright (c) 2024-2026 Paul Fox <pgf@foxharp.boston.ma.us>
SPDX-License-Identifier: BSD-2-Clause