v0.9.52

More fixed-format and indicator support, and several rules brought in
line with IBM i. Every change was checked against a real IBM i (PUB400).

- Data structures: a LIKEDS or LIKE subfield can have DIM, written before
  or after it. %SIZE of a data structure is its length in bytes (it was
  the C++ object's size), and CLEAR takes a subfield or an element.
- BASED data structures, free and fixed format, with LEN. The indicators
  are now held as IBM i holds them, '1' and '0', so a data structure based
  on %ADDR(*IN) gives them names. A pointer's INZ(%ADDR(x)) or INZ(*NULL)
  is no longer dropped.
- The *IN array: *IN(i), *IN = value, CLEAR *IN and %LOOKUP over *IN.
- SETON and SETOFF in fixed-format calculations.
- An indicator the program refers to but never sets is an error
  (RNF7030), as on IBM i; so is a fixed-format factor 1, factor 2 or
  result field that is not left-adjusted (RNF0262).
- The RPG cycle: a program whose calculations end with LR off runs them
  again, as IBM i does.
- A called program that returns with a halt indicator on is status 231 in
  its caller, and rpgc's own halt-indicator message carries RNX0233.

The IBM i execution run now builds the programs the tests call, so the
CALL tests are compared too: 225 tests give the same output on IBM i.
CTL-OPT USRPRF and VALIDATE are not planned; they are accepted and ignored.

v0.9.51

More of fixed-format RPG behaves as it does on IBM i, and two bugs found
by the October IBM Ideas portal review are fixed. Each change was checked
against a real IBM i (PUB400).

- I-spec record identification codes now take the zone (Z) and digit (D)
  parts of a position, not only the whole character (C). As on IBM i, '&'
  also matches zone C and '-' also matches zone D. A record belongs only
  to the first record type it matches; rpgc used to let a later type
  claim it too.
- *ENTRY PARM with factor 2 copies factor 2 into the parameter when the
  program returns, as IBM i does. rpgc used to refuse factor 2 there.
- LIKE(ds.field) defines a field like a qualified data structure
  subfield. It used to produce an integer field. A LIKE field now starts
  blank and takes INZ and an initial value.
- OVERLOAD now works when its candidates are prototypes declared
  elsewhere in the source (DCL-PR), not only procedures.

OpenRPG extensions IBM i does not accept: LIKE through a nested data
structure (ds.sub.field), and OVERLOAD telling an IND argument from a
character one.

v0.9.50

Two more behaviours brought in line with IBM i, each found or confirmed
by running programs on a real IBM i (PUB400).

- READ of a program-described file sets on the record identifying
  indicator of the record type it identified and turns no other one
  off; a READ at end of file changes none. rpgc used to turn off every
  other record type's indicator first. This was the cause of the last
  two unexplained differences on IBM i.
- MOVE or MOVEL of a character whose EBCDIC digit portion is above 9
  ('.', '+', '*', '<' and others) into a number is a decimal data error
  (status 907) that ends the program unless MONITOR handles it, and the
  result field is unchanged. rpgc used to set %STATUS, move zeros, and
  carry on.

Upgrading: a program that reads several record types with READ and
relied on rpgc turning the other indicators off must now set them off
itself before each READ, as on IBM i:

     C                   EVAL      *IN01 = *OFF
     C                   EVAL      *IN02 = *OFF
     C                   READ      CUSTFILE

A program that tested %STATUS after such a MOVE now ends at the MOVE;
put it in a MONITOR group.

The IBM i conformance runs log in to PUB400 once per run, retrying when
the machine closes a new connection, and carry on when another job's
output queue survives the library clear.

452 OpenRPG tests, 45 dspfc tests and 31 interactive display-file tests
pass.

v0.9.49

MOVE and MOVEL now behave as they do on IBM i. Each difference was
found by running the test programs on a real IBM i (PUB400), and each
fix was checked there.

- A negative number moved into a character field carries its sign in
  the zone of the last digit, as a zoned number does: -12345 moves as
  1234N, and a last digit of 0 as '}'. The sign used to be dropped.
- A character field moved into a number takes the digit portion of each
  character's EBCDIC (CCSID 37) code, so 'A1 ' moves as 110 with no
  error; a D zone in the rightmost character (J-R, '}') makes the result
  negative. Letters used to be status 907.
- A MOVE that cannot make a valid date, time or timestamp (status 112),
  or puts a date outside the result field's format range (114), now ends
  the program unless MONITOR handles it. It used to set %STATUS and
  continue.

Upgrading: a program that checked %STATUS after a failed date MOVE now
ends at the MOVE instead. Put the MOVE in a MONITOR group and test
%STATUS in ON-ERROR. MOVE has no (E) extender, on IBM i or here.

450 OpenRPG tests, 45 dspfc tests and 31 interactive display-file tests
pass.

v0.9.48

Twenty more features RPG developers asked IBM for -- the second tier of
the IBM Ideas portal's most-voted RPG requests -- and the standard RPG
several of them needed. Every change is checked on a real IBM i
(PUB400): the standard fixes run the same there, and the new features
are OpenRPG extensions that IBM i's compiler rejects.

- %SCANRPL now honours its start and length operands; it used to scan
  the whole source. An empty scan string is status 100, as on IBM i (it
  looped forever).
- OPTIONS(*TRIM) on a parameter is now applied: the argument arrives
  with leading and trailing blanks trimmed. It was parsed and ignored.
- OPTIONS(*OMIT) on a VALUE parameter is now an error (RNF3753), as on
  IBM i.
- Growing a DIM(*AUTO) array past its maximum is status 124, as on IBM
  i; it was 121.
- %CHAR of %XFOOT of a PACKED array has the array's decimal places
  (3.00, not 3.000000).

- %IF(condition : value if true : value if false) (93 votes).
- Keyword arguments: proc(name => 'Ada' : title => 'Dr.') (49).
- %SCANRPL with *FIRST or *LAST (49, and 18).
- OPTIONS(*UPPER) and OPTIONS(*LOWER) on parameters (48).
- CHAR with no length: a string of any length (44).
- INZ(%LIST(...)) on an array (41).
- %XFOOT, %MAXARR, %MINARR and IN over ds(*).subfield (39).
- %COMPCORR(ds1 : ds2 {: n}): compare data structures (38).
- DEPRECATED procedures, with compile-time warnings (35).
- %LOOKUP(%KDS(key) : ds(*)): search on several subfields (34).
- %REPEAT(string : count) (29).
- DEFAULT(value) for *NOPASS and *OMIT parameters (28).
- %PROGNAME (27).
- Binary literals: B'10100001' (26).
- FOR-EACH item IN array INDEX(i) (22).
- A comparison procedure for SORTA and %LOOKUPxx: %PADDR(proc) (22).
- WHEN-IS-NOT and WHEN-NOT-IN (22).
- *EMPTY (16).
- %TOHEX (or %HEX) and %FROMHEX (14).
- /MESSAGE {*WARNING | *ERROR} 'text' (12).

Each is described in docs/GUIDE.md. Source that uses them compiles only
with OpenRPG.

- SELECT with an operand, with WHEN-IS and WHEN-IN (IBM i 7.4/7.5).
- ds(*).subfield in SORTA and %LOOKUP.
- arr(*NEXT) on a DIM(*AUTO) array.
- %BITAND, %BITOR, %BITXOR and %BITNOT of character operands, byte by
  byte, and with more than two operands.
- A field or literal passed for an OPTIONS(*OMIT) parameter; only *OMIT
  itself could be.
- A procedure with no DCL-PR called before its DCL-PROC.

- The post-release smoke test runs the release's own tests, from its
  tag; it used to take main's newer ones.
- The IBM i conformance jobs stop after 60 and 90 minutes; one stalled
  on the runner for six hours.

449 OpenRPG tests, 45 dspfc tests and 31 interactive display-file tests
pass.

v0.9.47

Fixes for the four known issues in v0.9.46. Every fix is checked on a
real IBM i (PUB400): it gives the same diagnostics, and the programs
print the same output.

- A %SUBST start or length written as a number must fit the field's
  declared length, as on IBM i: %SUBST(s : 9 : 5) of a CHAR(10) is now a
  compile error (RNF0365; a start outside the field is RNF0364). This
  applies to %SUBST in expressions too. A start or length known only at
  run time is still checked then (status 100).

- %SUBST as an assignment target: %SUBST(s : 2 : 3) = 'XYZ' puts the
  value into that part of the field, padded with blanks or cut to fit,
  for a CHAR or VARCHAR field, an array element or a subfield, in free
  and fixed form.
- A name declared twice in one scope is reported by rpgc (RNF3316)
  instead of failing in the C++ compiler. A procedure's local may still
  hide a global of the same name.
- A call to a name that is no procedure, prototype or array is reported
  by rpgc (RNF7030) instead of failing in the C++ compiler.
- Fixed-format source may open with a compiler directive (/IF, /DEFINE,
  /COPY) or a /FREE block; it used to be read as free-format.

394 OpenRPG tests, 45 dspfc tests and 31 interactive display-file tests
pass.

v0.9.46

Twelve features RPG developers asked IBM for and IBM declined, from the
IBM Ideas portal's most-voted RPG requests, plus fixes for #26 to #29.
Every change is checked on a real IBM i (PUB400): the fixes run the
same there, and the new features are OpenRPG extensions that IBM i's
compiler rejects.

- INT(20) and UNS(20) are 64-bit, and every integer size is range
  checked (#26).
- A date plus or minus months or years keeps to the end of a shorter
  month: D'2024-01-31' + %MONTHS(1) is 2024-02-29, not 2024-03-02.
- EXEC and SQL must be on the same line, as on IBM i (the statement
  itself can still start on the next line).
- The SQLCA has no SQLERML; use SQLERRML or SQLERL, as on IBM i.

- Block comments: /* ... */, in free-form and fixed-form source (93 votes).
- Display files defined in JSON: `dspfc screen.json`, in the shape of
  the .dspfd descriptor (90 votes).
- %FKEY: the function key that ended the last EXFMT or READ (40 votes).
- Several names in one DCL-S: DCL-S i j k INT(10); (39 votes).
- Interpolated strings: f'I have {apples} apples' (37 votes).
- Procedures inside procedures, seeing the outer one's variables
  (30 votes).
- NOT IN (30 votes).
- Regular expressions: %MATCHES, %FIND, %COUNTMATCHES (28 votes).
- CTL-OPT OPTION(*NOLENCHK): a %SUBST length past the end gives the
  rest of the data (27 votes).
- %CHAR(number : *NOZEROSUPPRESS) keeps leading zeros (25 votes).
- AND, OR, NOT and parentheses in /IF and /ELSEIF (15 votes).
- *TRUE and *FALSE (13 votes).

Each is described in docs/GUIDE.md. Source that uses them compiles only
with OpenRPG.

- INZ(*ON) on an indicator starts it on (#27).
- Adding or subtracting a duration moves a timestamp, in every unit;
  subtracting a duration from a date or time compiles (#28).
- An indicator is character data, '1' or '0': 'flag=' + flag, c1 = flag
  and flag = '1' all compile (#29).
- x IN array compiles, for fixed and varying arrays.
- A procedure can return IND.
- In free-form source, an /ELSEIF after an /IF that was not taken is
  tested; its /ELSE used to run instead.

Found while working on this release, and not yet fixed:
- A name declared twice with DCL-S, and a call to a procedure that does
  not exist, are reported by the C++ compiler rather than by rpgc.
- %SUBST as the target of an assignment is a syntax error.
- Fixed-form source whose first line is a directive such as /IF is read
  as free-form.

383 OpenRPG tests, 45 dspfc tests and 31 interactive display-file tests
pass.

v0.9.45

Program calls now find the called program when they run, as on IBM i
(#23), and the runtime compiles without GCC warnings (#19). These changes
have not yet been checked on a real IBM i, because PUB400 is down.

Behaviour changes to be aware of

• Every program call (CALLP of an EXTPGM prototype, or CALL) is resolved
  at run time from a shared library. Callers no longer link a called
  program's object file. Build each called program with the new
  rpgc -shared ORD100.rpgle, which produces ORD100.so, ORD100.dylib or
  ORD100.dll, named after the source file, and put it next to the caller
  or on PATH. A called program written as a NOMAIN module needs to become
  a program: give it an *ENTRY PLIST, or DCL-PI *N in free form.

New

• Dynamic program calls (#23):
  • rpgc -shared builds a program that others call.
  • EXTPGM(variable) and CALL with a field in Factor 2 choose the program
    at run time. EXTPGM with no name means the prototype's own name.
  • The program is looked for in the caller's directory, then on PATH,
    then wherever the system loader looks. It is loaded on its first call.
  • A program that can't be found is status 211. One that ends in error is
    status 202 in its caller. CALLP(E) now reports an error's own status.
  • Parameters are passed in IBM i's formats (CHAR as bytes, INT as a
    native integer, PACKED as packed decimal, ZONED as digits), so a
    called program can be written in C, or be a GnuCOBOL module.
• Free-form programs can receive parameters, with DCL-PI *N among the
  main statements.

Fixes

• GCC no longer warns about the runtime on any compile (#19). Numbers are
  now formatted into strings sized to fit, which also fixes %CHAR of very
  wide decimal fields being cut short. The display-file runtime gets the
  same fix.
• Windows: an SQL program with a field named MSG compiles (also in v0.9.44).

Known issues

Found while testing this release, and not yet fixed: #26 (INT(20) is 32-bit),
#27 (INZ(*ON) on an indicator is ignored), #28 (adding a duration to a
timestamp leaves it unchanged), and #29 (an indicator can't be concatenated
to a character value).

349 OpenRPG tests, 39 dspfc tests and 28 interactive display-file tests pass.

v0.9.44

v0.9.43 plus one Windows fix. The v0.9.43 Windows x64 installer was not
built because its test suite stopped on this bug; this release includes it.

Fixes

• Windows: an SQL program with a field named MSG compiles. The SQL
  runtime's <windows.h> also declared the window manager's MSG structure,
  so the C++ compile failed with "reference to 'MSG' is ambiguous". This
  failure existed before v0.9.43; test335 exposed it.

For everything else in this release, see the v0.9.43 notes: fixes for the
issues reported against v0.9.42, commitment control for embedded SQL,
EXTFILE/OPEN/CLOSE, free-form DCL-F DISK(n), GET DIAGNOSTICS MESSAGE_TEXT,
halt indicators and the exit status.

v0.9.43

This release fixes the 24 issues reported against v0.9.42 (#1-#22, #24, #25).
These changes have not yet been checked on a real IBM i, because PUB400 is
down. A conformance run will follow when it is back.

Behaviour changes to be aware of

• Embedded SQL runs under commitment control, as CRTSQLRPGI's default
  COMMIT(*CHG) does on IBM i. Each change waits for COMMIT, and ROLLBACK
  now undoes it; before, every statement was committed as it ran.
  Changes still pending are committed when the program ends normally or
  at DISCONNECT, and rolled back when it ends in error. EXEC SQL SET OPTION COMMIT = *NONE; turns commitment control off (#2).
• SQLCODE is the code Db2 for i gives for the SQLSTATE, for example
  -204 for a missing table or -803 for a duplicate key, not the driver's
  native error number negated. An SQLSTATE with no Db2 equivalent keeps
  the native number (#22).
• A NULL fetched into a host variable that has no null indicator is
  SQLCODE -305 / SQLSTATE 22002, and the variable keeps its value (#25).
• A USROPN file is opened by OPEN. Before, USROPN was ignored and the
  file opened itself. Any other operation before OPEN is status 1211 (#20).
• In fixed-form source, a free-form statement outside /FREE (positions
  6-7 blank, code in 8-80) is now compiled. These lines used to be
  dropped without a message (#21).
• Calls are checked against the prototype (RNF7535). An argument of
  another type, or a non-variable passed to a parameter passed by
  reference, is now an rpgc error instead of a C++ compile error. A
  subfield of a CONST parameter can't be assigned (#17).

New

• EXTFILE('path') and EXTFILE(var) choose the file a program-described
  file opens. OPEN, CLOSE, CLOSE *ALL and %OPEN, with the E extender (#20).
• Free-form DCL-F name DISK(n) declares a program-described file, and
  USAGE(*INPUT:...) now parses (#21).
• GET DIAGNOSTICS CONDITION 1 with MESSAGE_TEXT, MESSAGE_LENGTH,
  RETURNED_SQLSTATE, DB2_RETURNED_SQLCODE and DB2_MESSAGE_ID, and the
  SQLCA's SQLERRMC and SQLERRML (#22).
• Exit status: halt indicators *H1-*H9 end the program in error with
  status n. rpg_set_exit_status (an OpenRPG extension called through an
  EXTPROC prototype) sets a status for a normal end (#24).
• Unnamed data structures: DCL-DS *N, and a fixed-form DS with a blank
  name (#15).
• A data structure can be used as a character value: DSPLY, assignment,
  concatenation, comparison and the string built-ins (#13).
• %INT, %DEC, %FLOAT and the other numeric conversions take a character
  operand (#8).
• Data structure subfields can be SQL host variables, written as
  :ds.sub or as :sub for a DS that isn't QUALIFIED (#6).

Fixes

• Fixed-form subfields that share From/To positions now share storage (#5).
• TEST(DE) takes a qualified subfield or an array element (#7).
• INZ(*SYS) on a fixed-form D-spec (#9).
• Subroutines in a subprocedure: an EXSR before the BEGSR now
  compiles (#10).
• OVERLAY subfields can be read through a LIKEDS parameter or a LIKEDS
  subfield (#16).
• EXEC SQL can be followed by a line break (#18).
• Fixed earlier, after v0.9.42 shipped: RETURN in a subroutine (#1);
  %EDITC and O-spec edit code lengths and alignment (#3, #4); a DS
  declared in a DCL-PROC (#11); SND-MSG *ESCAPE from the main procedure
  (#12); INZ on a DS or a subfield (#14). The g++ format-truncation
  warning (#19) came from code that was replaced, so it should be gone.

343 OpenRPG tests and 28 interactive display-file tests pass. A test can
now name its expected exit status in testN.exit, and error tests must be
rejected by rpgc itself, not by the C++ compiler.