Loading ...
Sorry, an error occurred while loading the content.

Re: mbyte.c patch

Expand Messages
  • Bram Moolenaar
    ... It seems defining FEAT_MBYTE_IME changes the method used for outputting DBCS text. That is not a good thing, deciding to use HanExtTextOut() should also
    Message 1 of 14 , Jul 15, 2002
    • 0 Attachment
      Glenn Maynard wrote:

      > On Sun, Jul 14, 2002 at 09:46:56PM +0200, Bram Moolenaar wrote:
      > > Setting 'encoding' to some Asian codepage should certainly work. Korean
      > > and Japanese users couldn't work without this.
      >
      > Aha. The HanExtTextOut code breaks this. If I comment that out, it
      > ends up coming to the "is_funky_dbcs" renderer, which renders correctly
      > via Unicode.

      It seems defining FEAT_MBYTE_IME changes the method used for outputting
      DBCS text. That is not a good thing, deciding to use HanExtTextOut()
      should also be based on runtime conditions. Perhaps checking
      "is_funky_dbcs" will help? Perhaps curwin->w_p_rl also needs to be
      checked? Hmm, HanExtTextOut() itself just calls ExtTextOut() when the
      character size is different from the "sysfixed" size. That suggests
      this check can be moved to before calling HanExtTextOut().

      > I don't see why they can't work without it--there should be no
      > functional difference if encoding=utf-8. Can you (or any CJK user
      > present) give an example? The "is_funky_dbcs" code should work fine in
      > all Win32. (The only thing that wouldn't work right now is the
      > alternate HanExtTextOut stuff, but that can be done in Unicode too, and
      > needs fixing anyway.)

      Don't trust what "should work" on MS-Windows! :-) Real user experience
      overrules everything else. Therefore every change must be tested in
      several conditions. Unfortunately, the code doesn't always say why it
      was needed...

      > > Unfortunately we can't drop all kinds of encodings and use UTF-8,
      > > conversion from/to Unicode will not always be possible.
      >
      > If there was no way to convert the input to Unicode, then there would be no
      > way to render it.
      >
      > The only two formats you can render in Windows are Unicode text and
      > ANSI-codepage text (ExtTextOutW and ExtTextOut, respectively)--and you
      > can always convert from the ANSI codepage to Unicode via
      > MultiByteToWideChar. MS has been helpful (for once) in making both of
      > these functions available in Win9x.

      But do they fully work? I don't know the details, but there are some
      problems with conversion from/to Unicode to/from MS-Windows codepages.
      You would lose information when using the clipboard, for example.

      > The only other part is file I/O--if iconv isn't available, MBToWC and
      > WCToMB need to be used when reading and writing files if the file
      > encoding is a codepage. I don't know if this is currently done.

      No, but it is one of the things to implement to avoid having to install
      iconv.

      > > There is the famous yen vs backslash problem, for example.
      >
      > Actually, that's not related.

      I'm not sure. I know quite a few people don't like iconv, because it
      does the conversion by the book instead of how the user expects it to
      be done.

      Anyway, the problem with conversions always is the risc of losing
      information. Unicode was designed to avoid that, but implementation
      problems may still cause trouble. Avoiding conversion avoids potential
      trouble. Perhaps I'm paranoid...

      > I notice there's a small hack in HanExtTextOut to render 0x5C as a
      > backslash in all DBCS locales, which only works in a specific font size.
      > This would need to be rewritten--but it needs to be rewritten anyway,
      > since it's 1: breaking "is_funky_dbcs", 2: kicks in when it shouldn't
      > (it doesn't apply to all DBCS), 3: can be done better, by flipping a
      > forward slash vertically, and 4: should be optional, as some Japanese
      > users will *want* 0x5C to be a Yen symbol. This is the same thing I
      > mentioned a while back, and something I still intend to implement
      > (port, really, from Putty). I'd also need to find out exactly when the
      > partially-composed-hangul code kicks in, since it's not happening with
      > the Win2K IME.

      We might as well do nothing. Most Japanese users are used to seeing the
      yen symbol in the place of a backslash anyway.

      --
      Team-building exercises come in many forms but they all trace their roots back
      to the prison system. In your typical team-building exercise the employees
      are subjected to a variety of unpleasant situations until they become either a
      cohesive team or a ring of car jackers.
      (Scott Adams - The Dilbert principle)

      /// Bram Moolenaar -- Bram@... -- http://www.moolenaar.net \\\
      /// Creator of Vim -- http://vim.sf.net -- ftp://ftp.vim.org/pub/vim \\\
      \\\ Project leader for A-A-P -- http://www.a-a-p.org ///
      \\\ Help me helping AIDS orphans in Uganda - http://iccf-holland.org ///
    • Glenn Maynard
      ... That s what I ended up doing. However, notice that one of the HanExtTextOut calls only does anything when ImeGetTempComposition returns non-NULL, and that
      Message 2 of 14 , Jul 15, 2002
      • 0 Attachment
        On Mon, Jul 15, 2002 at 10:32:44PM +0200, Bram Moolenaar wrote:
        > It seems defining FEAT_MBYTE_IME changes the method used for outputting
        > DBCS text. That is not a good thing, deciding to use HanExtTextOut()
        > should also be based on runtime conditions. Perhaps checking
        > "is_funky_dbcs" will help? Perhaps curwin->w_p_rl also needs to be
        > checked? Hmm, HanExtTextOut() itself just calls ExtTextOut() when the
        > character size is different from the "sysfixed" size. That suggests
        > this check can be moved to before calling HanExtTextOut().

        That's what I ended up doing.

        However, notice that one of the HanExtTextOut calls only does anything when
        ImeGetTempComposition returns non-NULL, and that only happens when
        bInComposition is TRUE.

        01:35am glenn@.../2 [~/vim/src] grep bInComposition *.c
        gui_w32.c:static BOOL bInComposition=FALSE;
        gui_w32.c: if (bInComposition == TRUE)

        It looks like this was broken at some point. It's apparently written to
        try to display a partially-composed Korean character with the Korean IME
        in Windows.

        I don't remember if this code predates the newer IME code; it might no
        longer be necessary, now that the IME is being set up correctly.
        (Perhaps the fact that it's not working and nobody's complained is
        evidence?)

        > We might as well do nothing. Most Japanese users are used to seeing the
        > yen symbol in the place of a backslash anyway.

        I view Japanese text in Vim daily, and I don't like seeing a backslash
        there.

        The backslash-fix code is a trivial addition to my font-repair code, which
        has other uses; in particular, fixing Unicode quotes in Japanese fonts.
        So I, at least, don't mind if the current backslash fix code goes away, since
        I have a replacement for it; and the current code is extremely special case
        (system-font-size only?), to the point of not really being useful.
        (And, like you said, if it *does* kick in, some users probably won't
        want it.)

        > But do they fully work? I don't know the details, but there are some
        > problems with conversion from/to Unicode to/from MS-Windows codepages.
        > You would lose information when using the clipboard, for example.

        If you convert with iconv you'll have problems. For example, if you edit
        a CP932 file by converting it with iconv on load, and then send something
        to the clipboard, and another application pastes that back in CP932 using
        the system conversions (MBtoWC), you'll lose data since the conversions
        aren't identical. Round-trip fails.

        The solution to that is to use the system's conversions instead of iconv
        for codepages. You're always using iconv if it's available (in fileio),
        and never using MultiByteToWideChar in Windows. I'd blame Unicode problems
        in Windows on that.

        The only way I can see system conversions causing problems is if round-trip
        conversion fails. I don't know of any cases of this, but if anyone does I'd
        definitely like to know.

        > Anyway, the problem with conversions always is the risc of losing
        > information. Unicode was designed to avoid that, but implementation
        > problems may still cause trouble. Avoiding conversion avoids potential
        > trouble. Perhaps I'm paranoid...

        > Don't trust what "should work" on MS-Windows! :-) Real user experience
        > overrules everything else. Therefore every change must be tested in
        > several conditions. Unfortunately, the code doesn't always say why it
        > was needed...

        Well, it's right to be paranoid in Windows. That's why we have this mailing
        list. :)

        I'll work on making fileio use MBtoWC for codepage<->Unicode conversion
        when possible. Even if you leave encodings as is, this will help.

        --
        Glenn Maynard
      • Bram Moolenaar
        ... This indeed looks very broken. ImeGetTempComposition() will always return NULL. It already was like this in 6.0. I didn t hear specific complaints about
        Message 3 of 14 , Jul 17, 2002
        • 0 Attachment
          Glenn Maynard wrote:

          > However, notice that one of the HanExtTextOut calls only does anything when
          > ImeGetTempComposition returns non-NULL, and that only happens when
          > bInComposition is TRUE.
          >
          > 01:35am glenn@.../2 [~/vim/src] grep bInComposition *.c
          > gui_w32.c:static BOOL bInComposition=FALSE;
          > gui_w32.c: if (bInComposition == TRUE)
          >
          > It looks like this was broken at some point. It's apparently written to
          > try to display a partially-composed Korean character with the Korean IME
          > in Windows.
          >
          > I don't remember if this code predates the newer IME code; it might no
          > longer be necessary, now that the IME is being set up correctly.
          > (Perhaps the fact that it's not working and nobody's complained is
          > evidence?)

          This indeed looks very broken. ImeGetTempComposition() will always
          return NULL. It already was like this in 6.0. I didn't hear specific
          complaints about this. I suppose this means we might as well remove
          this code.

          > > We might as well do nothing. Most Japanese users are used to seeing the
          > > yen symbol in the place of a backslash anyway.
          >
          > I view Japanese text in Vim daily, and I don't like seeing a backslash
          > there.

          Oh yes, the other way around it's not so nice.

          > The backslash-fix code is a trivial addition to my font-repair code, which
          > has other uses; in particular, fixing Unicode quotes in Japanese fonts.
          > So I, at least, don't mind if the current backslash fix code goes
          > away, since I have a replacement for it; and the current code is
          > extremely special case (system-font-size only?), to the point of not
          > really being useful. (And, like you said, if it *does* kick in, some
          > users probably won't want it.)

          It even uses floating point numbers. Let's get rid of this.

          > The solution to that is to use the system's conversions instead of iconv
          > for codepages. You're always using iconv if it's available (in fileio),
          > and never using MultiByteToWideChar in Windows. I'd blame Unicode problems
          > in Windows on that.
          >
          > The only way I can see system conversions causing problems is if round-trip
          > conversion fails. I don't know of any cases of this, but if anyone does I'd
          > definitely like to know.

          Perhaps it goes wrong when the codepage is set wrong? This is
          especially for 8-bit codepages where you probably don't notice the
          mistake if you do use the right font. Conversion to Unicode will reveal
          the problem.

          > I'll work on making fileio use MBtoWC for codepage<->Unicode conversion
          > when possible. Even if you leave encodings as is, this will help.

          That is a good thing. More people will be able to use different
          encodings "out of the box".

          --
          Engineers will go without food and hygiene for days to solve a problem.
          (Other times just because they forgot.)
          (Scott Adams - The Dilbert principle)

          /// Bram Moolenaar -- Bram@... -- http://www.moolenaar.net \\\
          /// Creator of Vim -- http://vim.sf.net -- ftp://ftp.vim.org/pub/vim \\\
          \\\ Project leader for A-A-P -- http://www.a-a-p.org ///
          \\\ Help me helping AIDS orphans in Uganda - http://iccf-holland.org ///
        • Glenn Maynard
          On Wed, Jul 17, 2002 at 12:54:07PM +0200, Bram Moolenaar wrote: This indeed looks very broken. ImeGetTempComposition() will always return NULL. It
          Message 4 of 14 , Jul 17, 2002
          • 0 Attachment
            On Wed, Jul 17, 2002 at 12:54:07PM +0200, Bram Moolenaar wrote:
            > This indeed looks very broken. ImeGetTempComposition() will always
            > return NULL. It already was like this in 6.0. I didn't hear specific
            > complaints about this. I suppose this means we might as well remove
            > this code.

            Alright. Patch attached:

            gui_w32.c:
            Remove HanExtTextOut, bInComposition, ImeGetTempComposition,
            DisplayCompStringOpaque.

            Removed a now-unneeded pair of braces from gui_mch_draw_string. I
            didn't unindent the block of code between it, since that'd bloat the
            patch with stuff that looks like changes but isn't. (I wish patch
            could handle this better.) I'll let you do that.

            Remaining problems in the renderer:

            SBCS encodings should render with "is_funky_dbcs"; right now we'll get
            garbage unless the encoding matches the system.

            The Arabic/Hebrew hack isn't always done. If we pass that range of text
            to the renderer as a block, we'll get weird results. Even if we don't
            have RL support enabled (or even compiled), and we're in Unicode, and the
            Hebrew text wouldn't look right to a Hebrew user, we still might be a
            regular user editing mixed-language text; we don't want weird things
            happening.

            len isn't set right in is_funky_dbcs; wide characters cause underlining
            glitches.

            Padding isn't set right in Unicode or is_funky_dbcs, so bold and italic
            characters cause spacing glitches.

            Also in this patch:

            gui_w32.c:
            Move padding computation to a function (set_padding).

            Move ETO_IGNORELANGUAGE setting out of each individual call, to
            make sure it's always used.

            Set padding in Unicode.

            (I'd have put these in a separate patch, but they're working on the same
            block of code.)

            This fixes bold/italics in Unicode, and simplifies the renderer a
            little.

            I think these problems are ultimately because there are too many code
            paths. I'd suggest always converting the string to Unicode at the start
            of the function. (At least in NT, it shouldn't be a speed hit, since I
            believe the font API will convert to Unicode anyway.) The RL (RevOut)
            special case needs to be done in Unicode before this will work. (Which
            I have code for, but am holding off on in the interests of keeping the
            patch down.)

            If this was done, the rest of the above problems would just go away.

            > > The only way I can see system conversions causing problems is if round-trip
            > > conversion fails. I don't know of any cases of this, but if anyone does I'd
            > > definitely like to know.
            >
            > Perhaps it goes wrong when the codepage is set wrong? This is
            > especially for 8-bit codepages where you probably don't notice the
            > mistake if you do use the right font. Conversion to Unicode will reveal
            > the problem.

            It'd cause problems with the clipboard, though.

            Hmm. I think the default encodings could be improved a bit. For
            example, if a user tries to load a file, and it fails, they're likely
            to first change "encoding". Of course, changing that alone isn't
            correct, but it may happen to work. Or, they may change just
            "fileencodings", which is closer, but that probably won't work, since
            you can't convert most other encodings to latin1. You have to change two
            values, and it takes a bit of reading to figure out exactly what to do.
            (Enough that a lot of users are likely to get it wrong.)

            First, I'd change the default "encodings" to UTF-8 in Windows. "latin1"
            is only a reasonable default if the system encoding happens to be one
            that's like latin1; UTF-8 is almost always a better default (especially
            since most users should be able to leave it alone.)

            Second, I'd change the default Unicode fencs from "ucs-bom,utf-8,latin1"
            to "ucs-bom,utf-8,CP####" in Windows. Again, latin1 is only a
            reasonable default for latin1 users; setting it intelligently should
            work for a lot more people without changes. (I think this should let
            most people edit locally-encoded files without having to touch encoding-
            related settings at all.)

            > > I'll work on making fileio use MBtoWC for codepage<->Unicode conversion
            > > when possible. Even if you leave encodings as is, this will help.

            Attached.

            This is straightforward, except for one thing: MBtoWC and WCtoMB are bad
            at handling streamed data. It has no way to nicely handle a final
            incomplete sequence.

            I've dealt with this by testing both; if converting the whole string
            doesn't work, bump the last character into the save buffer and try
            again. (It doesn't actually convert, so it shouldn't be too slow, but
            if it is I can optimize by doing this test manually.)

            Once this is tested, along with the clipboard, we might try asking any
            resident CJK people to try working in encoding=UTF-8 and tell us if they
            notice any problems (or if any problems are fixed.)

            That's probably the best way to see if using UTF-8 will cause problems:
            fix the known ones (which I'm doing) and then ask people to try it.
            Drop the theoreticals. :)

            --
            Glenn Maynard
          • Bram Moolenaar
            ... Ehm, this removes HanExtTextOut() completely. I m not sure if that is a good idea, it wan t added for fun. But it should certainly be called in a
            Message 5 of 14 , Jul 18, 2002
            • 0 Attachment
              Glenn Maynard wrote:

              > On Wed, Jul 17, 2002 at 12:54:07PM +0200, Bram Moolenaar wrote:
              > > This indeed looks very broken. ImeGetTempComposition() will always
              > > return NULL. It already was like this in 6.0. I didn't hear specific
              > > complaints about this. I suppose this means we might as well remove
              > > this code.
              >
              > Alright. Patch attached:
              >
              > gui_w32.c:
              > Remove HanExtTextOut, bInComposition, ImeGetTempComposition,
              > DisplayCompStringOpaque.
              >
              > Removed a now-unneeded pair of braces from gui_mch_draw_string. I
              > didn't unindent the block of code between it, since that'd bloat the
              > patch with stuff that looks like changes but isn't. (I wish patch
              > could handle this better.) I'll let you do that.

              Ehm, this removes HanExtTextOut() completely. I'm not sure if that is a
              good idea, it wan't added for fun. But it should certainly be called in
              a different way, and ImeGetTempComposition() can be deleted. The check
              for the sysfixed size should be done before calling HanExtTextOut(), and
              the call to ExtTextOut() inside HanExtTextOut() removed. This is a lot
              safer than deleting that code without knowing why it was there.

              > Remaining problems in the renderer:
              >
              > SBCS encodings should render with "is_funky_dbcs"; right now we'll get
              > garbage unless the encoding matches the system.

              I suppose the flag should be "non_system_codepage", which is set when
              using an 'encoding' different from the active codepage. I suppose it's
              not useful for Unicode values of 'encoding' though.

              I can't overview all the other remarks. I hope the people who worked on
              this code can respond. There is a lot of trial-and-error stuff in here,
              unfortunately.

              > I think these problems are ultimately because there are too many code
              > paths. I'd suggest always converting the string to Unicode at the start
              > of the function. (At least in NT, it shouldn't be a speed hit, since I
              > believe the font API will convert to Unicode anyway.) The RL (RevOut)
              > special case needs to be done in Unicode before this will work. (Which
              > I have code for, but am holding off on in the interests of keeping the
              > patch down.)

              That's quite a drastic change. I'm not really sure it's worth taking
              the risc.

              > > Perhaps it goes wrong when the codepage is set wrong? This is
              > > especially for 8-bit codepages where you probably don't notice the
              > > mistake if you do use the right font. Conversion to Unicode will reveal
              > > the problem.
              >
              > It'd cause problems with the clipboard, though.

              Unless the program on the other side has the same (reverse) problem.

              > Hmm. I think the default encodings could be improved a bit. For
              > example, if a user tries to load a file, and it fails, they're likely
              > to first change "encoding". Of course, changing that alone isn't
              > correct, but it may happen to work. Or, they may change just
              > "fileencodings", which is closer, but that probably won't work, since
              > you can't convert most other encodings to latin1. You have to change two
              > values, and it takes a bit of reading to figure out exactly what to do.
              > (Enough that a lot of users are likely to get it wrong.)

              Changing 'encoding' should be discouraged, because it invalidates text
              in other buffers, registers, etc. Best is to use something like:

              :edit ++enc=cp123 file

              Setting 'fileencodings' to a good value would also work. This mostly
              requires a Unicode value for 'encoding'.

              > First, I'd change the default "encodings" to UTF-8 in Windows. "latin1"
              > is only a reasonable default if the system encoding happens to be one
              > that's like latin1; UTF-8 is almost always a better default (especially
              > since most users should be able to leave it alone.)

              The problem with using "utf-8" as a default is that a new file will get
              this encoding, which is probably not what people expect. I think a new
              file should probably use the active codepage as a default, unless the
              user has selected something else.

              Also don't forget that what we call "latin1" is actually any 8-bit
              encoding. It doesn't have to be the right name, it only matters when
              doing conversions. Quite often we don't actually know what encoding is
              being used and fall back to using latin1. This applies more to Unix
              than MS-Windows though, since MS-Windows supplies a function to obtain
              the active codepage.

              > Second, I'd change the default Unicode fencs from "ucs-bom,utf-8,latin1"
              > to "ucs-bom,utf-8,CP####" in Windows. Again, latin1 is only a
              > reasonable default for latin1 users; setting it intelligently should
              > work for a lot more people without changes. (I think this should let
              > most people edit locally-encoded files without having to touch encoding-
              > related settings at all.)

              Same issue. Instead of "latin1" the active codepage could be used, if
              it is an 8-bit encoding. Otherwise it must be "latin1", because we
              always need to fall back to an 8-bit encoding.

              > > > I'll work on making fileio use MBtoWC for codepage<->Unicode conversion
              > > > when possible. Even if you leave encodings as is, this will help.
              >
              > Attached.

              I'll look into this later.

              --
              You have heard the saying that if you put a thousand monkeys in a room with a
              thousand typewriters and waited long enough, eventually you would have a room
              full of dead monkeys.
              (Scott Adams - The Dilbert principle)

              /// Bram Moolenaar -- Bram@... -- http://www.moolenaar.net \\\
              /// Creator of Vim -- http://vim.sf.net -- ftp://ftp.vim.org/pub/vim \\\
              \\\ Project leader for A-A-P -- http://www.a-a-p.org ///
              \\\ Help me helping AIDS orphans in Uganda - http://iccf-holland.org ///
            • Glenn Maynard
              ... Well, there are two parts to it: One s the fake-backslash code. I don t think that s worth keeping as is, since it s very special-case, and like you said,
              Message 6 of 14 , Jul 18, 2002
              • 0 Attachment
                On Thu, Jul 18, 2002 at 10:02:00PM +0200, Bram Moolenaar wrote:
                > Ehm, this removes HanExtTextOut() completely. I'm not sure if that is a
                > good idea, it wan't added for fun. But it should certainly be called in
                > a different way, and ImeGetTempComposition() can be deleted. The check
                > for the sysfixed size should be done before calling HanExtTextOut(), and
                > the call to ExtTextOut() inside HanExtTextOut() removed. This is a lot
                > safer than deleting that code without knowing why it was there.

                Well, there are two parts to it:

                One's the fake-backslash code. I don't think that's worth keeping as
                is, since it's very special-case, and like you said, most Japanese users
                don't mind (expect, actually) the Yen sign. (I do have a replacement,
                since *I* mind the Yen sign, but it's not urgent.)

                The important one is the Korean IME code. I'm not sure if that's needed at
                all. I'm *guessing* that originally, the partially-composed characters
                weren't being displayed, and this was hacked in. However, it's wrong; the
                right way is to tell the IME where the cursor is so it renders it. This is
                being done now, due to the IME patch, so I believe this is completely obsolete.

                Of course, that's a guess. Could we contact whoever wrote this patch
                and ask them if the current Vim release is working for him in the IME?
                (Searching the archive for Sung-Hoon Baek didn't help.)

                > I suppose the flag should be "non_system_codepage", which is set when
                > using an 'encoding' different from the active codepage. I suppose it's
                > not useful for Unicode values of 'encoding' though.

                I'd say "not_system_codepage". (Unicode is not a system codepage; it's not
                a non-system-codepage.)

                > > I think these problems are ultimately because there are too many code
                > > paths. I'd suggest always converting the string to Unicode at the start
                > > of the function. (At least in NT, it shouldn't be a speed hit, since I
                >
                > That's quite a drastic change. I'm not really sure it's worth taking
                > the risc.

                It could be done piecemeal over a number of releases. However, if
                encoding can default to UTF-8, then all that's really important is
                getting UTF-8 rendering right in all cases, and that's something that
                should be done anyway.

                > The problem with using "utf-8" as a default is that a new file will get
                > this encoding, which is probably not what people expect. I think a new
                > file should probably use the active codepage as a default, unless the
                > user has selected something else.

                New files get the "fileencoding" encoding, right? Make fileencoding do
                this, too, then.

                > Also don't forget that what we call "latin1" is actually any 8-bit
                > encoding. It doesn't have to be the right name, it only matters when
                > doing conversions. Quite often we don't actually know what encoding is
                > being used and fall back to using latin1. This applies more to Unix
                > than MS-Windows though, since MS-Windows supplies a function to obtain
                > the active codepage.

                However, conversions are important. It's pretty confusing if Vim is
                appearing to load and display a file correctly, but sends junk to the
                clipboard because the conversion was done incorrectly.

                > Same issue. Instead of "latin1" the active codepage could be used, if
                > it is an 8-bit encoding. Otherwise it must be "latin1", because we
                > always need to fall back to an 8-bit encoding.

                What about fencs="ucs-bom,utf-8,CP####,latin1"?

                --
                Glenn Maynard
              Your message has been successfully submitted and would be delivered to recipients shortly.