gcc-log.txt

=================================================
Combined gSTLFilt.pl / Proxy c++.cpp Revision Log
=================================================

====
2008
====

2/3
    Template metaprogramming sample metatest.cpp was updated to reflect
    changes to MPL header file and template names.

1/30 
    Modified the gfilt shell script a bit. It now expects gSTLFilt.pl to 
    reside in the same dirctory gfilt itself was found in.
    Also, the usage message has been expanded a bit to make gfilt's role
    clearer. Thanks to Pedro Estrela for the gfilt suggestions.

1/28 (gSTLFilt.pl v3.10)
    Well, I guess MinGW 3.4.5 wasn't as new as some other releases. I've
    now added support for MinGW gcc 4.x, tested under MinGW gcc 4.1.1 and
    also under TDM gcc 4.2.2. 

1/23 (gSTLFilt.pl v3.00)
    FINALLY updated the Perl script to provide full functionality for the
    the latest gcc compiler releases. I've tested using MinGW 3.4.5.
    While updating the filter for the latest gcc generation, several issues
    with previous versions of gcc have been resolved as well. list, deque,
    and associative container iterators are filtered more precisely than
    they were before.

    In addition to the "gen_set" (set or multiset) and gen_map (map or
    multimap) names, there's now a new one: "gen_set_or_map", which could
    be any of the four standard containers. Sorry, but some cases all four
    containers' iterators reduce to the identical iterator type and therefore
    cannot be distinguished from each other.

    On the bright side, any year now gcc will begin to support "concepts" and
    you'll finally be able to throw STLFilt away, lol. The sooner the better!



====
2005
====

6/29 (gSTLFilt.pl v2.47a)
    Fixed problem with $def_iter_policy being set to "L" not working right
    (thanks, Dave Abrahams, for pointing this one out.)

    Also added more comments and improved $COPTS handling for gfilt (the
    Unix shell script driver)

2/16
    A bit more cleanup to the gfilt script. It now works on freeBSD (added
    line-continuation to the long list of cases in the case statement...
    thank you Simon Barner for pointing that one out). 
    Also removed a spurious case from the case list. 

2/11
    The 'gfilt' script (for *nix distributions) was accidentally packaged
    into the .tar-format distribution with Windows-style (CRLF) line
    terminators, causing some shell interpreters grief. This has been
    corrected.


====
2004
====


12/17
    Modified the *nix shell script driver "gfilt". It had been missing
    support for several new options that had been added to the Perl script.
    Also added some more comments to gfilt so it would contain at least some
    rudimentary documentation.

11/26 (gSTLFilt.pl v2.47)
    Added the ability to strip trailing "null"-style boost arguments
    from template argument lists. By default, the arguments
       boost::tuples::null_type
    and
       mpl_::na
    are removed. The following notes all relate to this feature.

    Any number of additional fully-qualified type names may be added by
    specifying the -nullarg:xxx option once for each type name (with xxx
    being the name to be deleted).

    if xxx is 'clear', the null argument list is emptied; subsequent -nullarg
    specifications appearing on the command line will work, but not the
    defaults as listed above (unless they're re-specified using the -nullarg
    option later on the command line.)

    The default list of automatically-deleted typenames may be configured by
    editing the @nullargs variable in the user-configurable section of the
    Perl script.

11/25 (gSTLFilt.pl v2.46)
    Corrected filtering bug related to std::vector.

    Introduced the "null"-style argument stripping feature, but decided to
    change the way it works a bit. See the description for the v2.47 release
    above.

08/29 (gSTLFilt.pl v2.45)
    Patched a bug that caused in internal error. You don't want to know
    the details...and you especially wouldn't want to know the fix ;-)

08/10 (gSTLFilt.pl v2.44)
    Corrected a bug that broke the "with clause" substitution routine when
    an array dimension value in square brackets appeared within the
    substituted types. 

2/21 (gSTLFilt.pl v2.43a)
    Augmented list of recognized source file extensions within the Perl
    script for consistency with recent c++.cpp corrections.

1/16 (gSTLFilt.pl v2.43)
    Corrected a bug where template instantiation backtrace re-ordering would
    fail to emit the backtrace at all in certain cases when the backtrace
    appeared at the very end of the error stream.

1/15 (c++.cpp v3.43a)
    Added ".cc" as a recognized C++ source file extension, and ".o" as a
    recognized object file extension, to the Proxy compiler.
    Thanks to Stefan Naewe for pointing out these omissions.


====
2003
====

12/24
    Fixed inconsistencies between code, comments and documentaion with respect
    to some default values of the Proxy-gcc.INI configuration parameters.

12/23 (gSTLFilt.pl v2.42, c++.cpp v3.43)
    Corrected a /meta:D mode parsing bug that caused internal errors if the
    sequence ">> " appeared in a message in the context of closing delimiters
    for nested template parameter lists.

    Corrected another /meta:D bug involving appearances of the sequence "()"
    in various contexts.

    Corrected a premature closing of the NativeLog.txt file in the event of
    an internal parsing error, causing errors on all subsequent attempts to
    write to the file.

    The Proxy c++.exe now accepts the /lognative option to force creation of
    the Perl script debugging log file NativeLog.txt. Formerly, the only way to
    specify creation of the log file when using the Proxy c++ was to configure
    the option LOG_NATIVE_MSGS in Proxy-gcc.INI.

12/17 (gSTLFilt.pl v2.41a [really!], updated Proxy-gcc.INI):
    Oops, the v2.41a Perl script didn't make it into the distributions from
    12/04 as it was supposed to. Sorry about that.

    Several corrections to the comments in Proxy-gcc.INI:

    1. The line of the [common] section of Proxy-gcc.INI that began with:

            ;native_cl=c++2.exe
       
       incorrectly used the name "native_cl" for the option. It should have
       begun with:

            ;native_gcc=c++2.exe

       The remaining commentary on that line was also incorrect and has now
       been fixed and clarified.

    2. Correction: In the comment at the end of the DEBUG_DIR line, the debug
       log file name created automatically is gcc-dblog.txt, *not* cl-log.txt.

12/04 (gSTLFilt.pl v2.41a)
    Fixed wrapping error in /break:D mode when the text ">", "<", ">>"
    or "<<" appeared in a message.

11/24 (gSTLFilt.pl v2.41)
    Corrected a problem where "within this context" lines were getting
    separated from the lines they referred to.

11/19 (gSTLFilt.pl v2.40)
    Made sure that lines are never wrapped in the middle of pathnames, even
    if that means an output line length may exceed the /width setting (figuring
    it is more important for editors to be able to automatically locate the
    point of the error than it is to unconditionally honor the /width value).

11/14 (gSTLFilt.pl v2.39e)
    In messages referring to object files, gcc's mangled-names-from-hell are
    sometimes followed on the same line by long typenames that really need
    wrapping. Those long typenames are now processed instead of being "thrown
    out with the bath water" ;-)

11/12 (gSTLFilt.pl v2.39d)
    Patched an obscure condition that drew a
        "Use of uninitialized value in multiplication (*)..."
    error during -break:D mode wrapping. 

11/11 (gSTLFilt.pl v2.39c)
    Corrected Dave-mode reordering bug where multi-line primary error messages
    were being split up, with the instantiation backtrace ending up in the
    middle.

    Corrected possible problem recognizing "const pointers" (as opposed
    to pointers-to-const) in certain constructs.

08/08 (gSTLFilt.pl v2.39a)
    Tuned some of the comments at the top of the script.

07/29 (gSTLFilt.pl v2.39)
    Added two user-configurable variables to the Perl script (no command-
    line access to these):
        $keep_stdns:  if 1, does NOT strip the standard namespace qualifiers
                      "std::", "__gnu_cxx", "_STL" or "_STLD" from names.
                      Note that the usual behavior is to strip them, and all
                      the STL library error filtering counts on that having 
                      been done. This option, when set to 1, is useful only to
                      retain all information in conjunction with long typename 
                      wrapping (-break:D). 

        $show_internal_err:
                      1 to display information about internal errors in the 
                      long typename wrapping (-break:D) mechanism, 0 to
                      suppress them. Please leave this at 1 and follow the
                      instructions re. emailing the log file if an error
                      occurs. If you need to set it to 0 due to a persistent
                      error, please set it back to 1 after a fix has been
                      provided. Thanks!
                         

06/06 (gSTLFilt.pl v2.38)
    Corrected Plain-mode wrapping logic to fix another off-by-1 error that
    had been emitting spurious blank lines.

06/05 (c++.CPP v3.42a)
    Wildcards (such as *.cpp) now work on the Proxy c++ command line (just
    needed to be linked with SETARGV.OBJ).

05/31 (gSTLFilt.pl v2.37)
    /iter:L now forces $newiter to 'iterator', so the word 'iterator' is
    never shortened in /iter:L mode.

05/29 (gSTLFilt.pl v2.36)
    Several /break:D mode bugs fixed:
       - '(' character with spaces around it was not being recognized as in-
          creasing the nesting level, causing an internal error later on the
          line when the matching ')' was encountered
       -  '>=' and '<=' operators were not excluded from '>' and '<' detection
       -  indentation was off by a space during /closewrap processing

    Dropped the $space_close_brackets option; it now always puts spaces between
    consecutive '>' characters (unless they're the >> operator, with any luck.)
    This was necessary to keep the /break:D wrapping logic from getting out of
    control.

    Improved detection of object/library filenames in messages (suffixes of .a,
    .o, .obj) so as to disable /break:D wrapping for those messages. The main
    problem was that certain mangled-names-from-hell were wreaking havoc with
    the wrapping algorithm.

05/15 (gSTLFilt.pl v2.35, C++.CPP v3.42)
    Added new /meta option as a convenience to auto-configure the auto-wrapping
    options. See the entry from 2/21 below for details on the /break, /cbreak
    and /closewrap options related to this:

    /meta:x       Configure for metaprogramming [Y] or not [N] as follows:

                  /meta:y (or just /meta) forces /break:D, and forces /cbreak
                  and /closewrap options according to values specified in the
                  $meta_y_cbreak and $meta_y_closewrap variable initiali-
                  zations, respectively.

                  /meta:n forces /break:P (/cbreak and /closewrap don't apply.)

                  If no /meta option is present, the default values for
                  /break, /cbreak and /closewrap are determined by
                  the INI file (if using a Proxy compiler) or by the settings
                  of $break_algorithm, $comma_wrap and $close_wrap in the 
                  user-configurable area of the Perl script.

                  In either case, if the output width hasn't yet been set to
                  a non-zero value, it is set to 80 (choosing a wrapping
                  flavor makes no sense with wrapping disabled.) This may be
                  overridden by a subsequent /width option.

                  Note also that /meta may not be specified in the INI files,
                  because it is intended as a command-line "override"
                  mechanism. The individual settings of /break, /cbreak and
                  /closewrap, however, may be (and apply only when the /meta
                  option is not used.)
    
    A new Perl script user-configurable variable, $space_after_commas, forces
    a space after any comma appearing in messages. Default value is 0 (don't).

    The default output width is now 80 universally, including when the Proxy
    c++ is used. Formerly, the default was 80 in the Perl script but 0 when
    using the Proxy c++ ... this was inconsistent with the default /break:D
    mode setting.


04/22 (gSTLFilt.pl v2.34)
    Shortened the primary regex ($t) a tad more.

04/11
    STLTask (.EXE and sources) have been updated with a fix to the wxWindows
    library that keeps the tray icon from disappearing as a result of the
    taskbar re-initializing itself (such as after a Windows Explorer crash).
    Many thanks to Graeme Prentice for solving this long-standing problem.

04/08 (gSTLFilt.pl v2.33)
    Fixed a bug in STLTask that prevented it from recognizing the NATIVE_GCC
    configuration parameter in Proxy-GCC.INI (it was expecting "NATIVE_CL")

    Adjusted handling of the phrase 'iterator' to avoid shortening it unless
    one of the standard iterator types is involved, and added some comments
    to the user-configurable section to show how to disable all iterator
    filtering.

03/22 (gSTLFilt.pl v2.32)
    Fixed bug preventing "with clause" processing from handling more than
    one "with clause" per error message in certain cases.

    Fixed bug that caused an internal error in some cases when the "->"
    operator appeared in a message.

    Rather than just bailing out when an internal error occurs, the
    script now emits the unfiltered messages and resumes processing with
    the next line of input. Then at the END of the run, it requests that you
    send me the NativeLog.txt log file automatically created.

03/21 (C++.CPP v3.41)
    Fixed some messages that used the term "Proxy CL" instead of
    "Proxy c++"

03/19 (C++.CPP v3.40)
    The Proxy c++ now accepts quotation marks around pathnames in 
    Proxy-gcc.INI, and detects faulty quotation.

03/18
    Added simultaneous multi-platform support to STLTask via a new config-
    uration file, STLTask.INI, to be placed into the same directory as
    Proxy-gcc.INI. The new version of STLTask is backwards-compatible with the
    old Proxy-gcc.INI, so the new features don't show up unless you actually
    tailor and install STLTask.INI into the Windows directory. For more
    details, see the latest version of README-STLTask.txt in the package.

03/13 (c++.cpp v3.30)
    Proxy c++ command-line processing revamped considerably, to allow
    the C++ filename to be anywhere on the line (the previous logic, a
    hold-over from the MSVC Proxy CL, assumed the source file name is
    at or near the end of the command line and failed to launch Perl at
    all otherwise.)
    Main benefit: The Proxy c++ is now compatible with Dev-C++ !!

    Cleaned up QUICKSTART.txt a bit.

03/07 (c++.cpp v3.21)
    The name of the Proxy-gcc.INI option to configure the native compiler
    name is supposed to be "native_gcc", NOT "native_cl" as shown in the
    file as distributed. A big "Thank you!" goes to bdsoft BBS poster 
    "rost" for catching this. I've also scoured the documentation for
    related gaffes, and caught a few places where the term "Proxy CL" is
    used in the gcc context. I've changed them to read "Proxy compiler".

    [Note: the functionality of the c++.exe executable as distributed has
    not changed; differences are at the source code level only...well,
    and in the version number it prints out, but that's it.]

    The Unix shell script, gfilt, now appends the value of environment
    variable $GFILTOPTS, if it exists, to the internal variable $DOPTS as
    the documentation says it does. 

    I've documented, within some of the doc files, which sections apply to
    just Windows and which apply to both Windows and Unix.


03/03 (gSTLFilt.pl v2.31)
    Enhanced the /break:d mode algorithm to break before open quotes (`)
    if they have no matching close quote on the same line.

    Corrected a bug that caused breaks in the middle of words/names in certain
    cases when the line length was exactly equal to the width setting.

    Added support for re-ordering of #include file backtraces in /hdr:LD1 
    and /hdr:LD2 modes (previously only template instantiation backtraces
    were being re-ordered.)

    Overhauled the algorithm behind Standard Lib pathname-shortening (the
    /path:x command line option) considerably, for consistency and correct
    handling of std header names lacking extensions.


02/21 (gSTLFilt.pl v2.30)
    Added new line-wrapping logic designed specifically to aid readability
    of super-long typenames typical of template metaprogramming applications.
    (If you have Boost installed, the sample source file metatest.cpp
    illustrates these features.)

    New command-line options (these apply only when output width is non-zero):
        
        -break:x    Break algorithm: D[ave Abrahams] (default) or P[lain]
                    The "Dave" option wraps long complex typenames in a way
                    that makes it easier to see parameter lists at various
                    nesting depths.

        -cbreak:x   Comma break: B = break before commas (default), A = break
                    after. Just a matter of personal taste; Dave prefers
                    before, so that's the default.
            
        -closewrap:x    Wrap before unmatched close "paren" (paired delim):
                        Y (default), don't wrap: N 

    New script-only user-configurable variables:

        $space_close_brackets = 1;  # 1 to force spaces between consecutive >'s

        $wrap_own_msgs = 0;         # wrap STL Decryptor's status messages
                                    # to same output width as errors (1) or
                                    # don't (0)
        $smush_amps_and_stars = 0;  # insert (0) or don't insert (1) space
                                    # between typename and * or &

    Updated the gfilt shell script (for Unix) to support the new gSTLFilt.pl
    command-line options, and also corrected several issues in that script
    reported by George Katsirelos:
        1) Detection of gcc3 no longer improperly sends gcc version information
           to stdout
        2) The exit status of the entire script now properly represents the
           status of the gcc compilation (so that gfilt may be used from an-
           other script with proper conditional behavior)
        3) the PAGER variable may be set to blank to prevent use of any
           paginator

    (c++.cpp v3.20)
    New Proxy-gcc.INI option in [proxy.gcc] section:
        gcc2=true or false
    if true, suppresses the -fmessage-length=0 gcc command-line option. If
    false, supplies the option. If you switch between gcc2 and gcc3, I
    recommend setting gcc2 to true, then manually supply the
    -fmessage-length=0 option to gcc3 when necessary (e.g. via batch files 
    that are either gcc2-specific or gcc3-specific).


02/05 (gSTLFilt.pl v2.21)
    Optimized the primary regex, $t, to avoid "regex too long" blowups under
    some older versions of Perl. This should also result in substantially
    faster execution under all Perl implementations.

01/18 (gSTLFilt.pl v2.20, C++.CPP v3.10)
    Thanks to careful detective work and some new defensive coding above and
    beyond the call of duty by Thomas Becker, the package is now compatible
    with ActivePerl 5.8. 


====
2002
====

12/24 (gSTLFilt.pl v2.14)
    Corrected an off-by-1 error in handling of the /width:nn option,
    to allow nn to be the actual number of available columns.

    Made a few other adjustments to the line break algorithm.

12/19 (gSTLFilt.pl v2.13)
    Found more /cand:L and /hdr:L issues and cleaned them up.

12/18 (gSTLFilt.pl v2.12)
    Corrected "with" clause processing for certain errors generated while
    using "long" candidate and/or header policies, esp. with the hash-
    based containers.

    Added /with:x option to Perl script (and Proxy c++): If set to L or N,
    suppresses "with" clause substitutions, so you can see the original
    template parameter identifier names. If set to S or Y (the
    default), normal (to STLFilt) "with" clause substitution is performed.
    Thanks to Dave Abrahams for suggesting this option.

12/17 (gSTLFilt.pl v2.11)
    Added /hdr:LD and /hdr:LD2 header policy variations, suggested by
    Dave Abrahams:
        /hdr:LD[1] (long, Dave Abrahams form 1. LD and LD1 are identical):
            actual error msg moves to front of instantiation backtrace
        /hdr:LD2   (long, Dave Abrahams form 2):
            as above, with trigger line duplicated before the backtrace

    Added user-configurable Perl option, $reformat_linenumbers. If set to
    1, line numbers are transformed to an MSVC-like format. This may confuse
    some error-locating IDEs. Defaults to 0, retaining gcc's format.

10/20 (STLTask.cpp v2.6g)
    Fixed several bugs in STLTask: infinite dialog-box looping
    upon detection of corrupt configuration in certain cases, removed
    clipboard-related functionality (it was only tuned for the MSVC IDE),
    improved several status messages.

9/3 (gSTLFilt.pl v2.02, updated gfilt Unix shell script)
    The gfilt shell script (for Unix only) now distinguishes Decryptor
    options from compiler options, and dispatches each to the appropriate
    program.

    Minor tweaking of set iterator processing.

8/27 (gSTLFilt.pl v2.01)
    Fixed bug that emitted a spurious ":" toward end of Decryption output
    
    Corrected default initialization of the $banner variable, and associated
    comments. $banner is set to 'Y' now so the sign-on message is displayed
    by default. When invoked indirectly via the Proxy c++, the Proxy c++
    includes the /banner:N option to the script to turn the sign-on message
    off (because the proxy c++ itself has by then already displayed a 
    sufficient sign-on banner).
 
    (C++.CPP v3.01)
    Now forces "/banner:N" option to suppress gSTLFilt.pl's sign-on banner.

    QUICKSTART.txt: Corrected example of "trivial command line usage" to
    include the standard error redirection clause "2>&1" (necessary with
    gcc so that the diagnostics, emitted on the standard *error* stream,
    are fed properly to Perl for Decryption).

8/25 (gSTLFilt.pl v2.00)
    Support for gcc 3.1 added. Same script now filters both 2.95.x and 3.x.

    (C++.CPP v3.0)
    Initial release of gcc-specific Proxy compiler, c++.exe, for Windows
    distribution only (source code included as c++.cpp)


- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

4/11 (gSTLFilt.pl v.99)
    Removed unnecessary regex substitutions still present after porting from
    previous versions of the Decryptor.

3/12 (gSTLFilt.pl v.98)
    Added a new variation of the header policy option: /hdr:nn, where nn is a
    number, sets an internal variable named $headers_to_show so that nn
    messages are passed thorough from each cluster of standard header messages,
    and the rest are suppressed.
    You may change the default value of $headers_to_show by setting the value
    of $def_hdrs_to_show in the user-configurable section of gSTLFilt.pl,
    and the new value will apply any time you use /hdr:M.
    Note: Since the /hdr:M option automatically sets $headers_to_show to the
    value of $def_hdrs_to_show, do not use /hdr:M and /hdr:nn simultaneously.

    Whenever headers or candidates are suppressed, the filter ends its session
    by reminding the user to use /hdr:L and/or /cand:L (as appropriate) to get
    more detail if needed. [If you get sick of seeing the reminders, just set
    the user-configurable variable "$advise_re_policy_opts" to 0.]

    The Windows batch script gfilt.bat, and unix driver script gfilt, now look
    in the environment for a variable named GFILTOPTS. If it is present, its
    value overrides the setting of OPTIONS built in to the script. This allows
    the user to try different filtering options without having to edit the
    script each time or manually enter the entire command pipeline sequence.

    Updated README.txt to properly describe behavior of /hdr:x and /cand:x.

3/11 (gSTLFilt.pl v.97)
    Corrected another bug where multisets were being incorrectly translated to
    sets. 
    Now strips the {} from {anonymous} so that "anonymous" is recognized as a
    type name.
    Added some documentation to the Bugs/Limitations section of README.txt to
    explain my "generic" iterator approach to set/multiset iterators and map/
    multimap iterators.

3/10 (gSTLFilt.pl v.96)
    Corrected several filtering bugs w/respect to map/multimap iterators,
    where these iterators were incorrectly transformed into *set* iterators 
    and/or their const-ness was attributed incorrectly.

2/26 (gSTLFilt.pl .95)
    Output line wrapping has been cleaned up a bit, so that all messages
    (with the exception of command-line parameter errors) are consistently
    wrapped to the desired column.

    Many miscellaneous Perl coding errors were corrected [next time, I've
    got to remember to run perl -W *before* releasing a script! ;-) ]

2/23 (gSTLFilt.pl .94)
    Added /lognative option (already had it in other version, somehow it was
    inadvertently omitted from the g++ version until now). When specified, this
    option causes the unfiltered g++ output to be written to a file with the
    named NativeLog.txt (name is hardwired, but of course may be changed in the
    Perl script). This assists in debugging the decryptor; if you need submit
    examples of errors that are not correctly filtered, using this option
    and capturing the output of the filter at the same time guarantees that the
    raw and filtered messages are in sync.

2/22 (gSTLFilt.pl .93)
    Added recognition of basic_string template instantiations in multithreaded
    configurations (first template param to __default_alloc_template true).
    Thanks to Marcin Kasperski for first reporting this problem.

2/20 (gSTLFilt.pl .92)
    Added /banner:x command line option to control ID banner. By default, it
    displays. To turn off the banner, use /banner:n

2/17 (gSTLFilt.pl .91)
    Major revision to the Perl script. Now includes the ability to suppress
    consecutive messages originating from within the STL standard headers.
    The script option /hdr:x, where x=S,M or L, controls header message 
    suppression:
        L: pass all messages through
        M: pass only first of a consecutive batch, and announces how many were
           suppressed
        S: suppress all, and announce how many were suppressed

    The /cand:x option works similarly for "candidate" lines (except that it is
    totally silent in 'S' mode, instead of reporting # of candidates
    suppressed).

    Fixed up many small issues regarding pathname processing and detection of
    standard header messages under Unix (used to only look for *back*-slashes.
    Oops.)


2/16 (gSTLFilt.pl 0.90)
    Initial beta release.