Re: IMPORTANT: proposed removal of non-bidi code

From: Tomas Frydrych (tomas@frydrych.uklinux.net)
Date: Thu May 09 2002 - 05:14:15 EDT

  • Next message: John L. Clark: "Re: AbiWord 1.0.1 released."

    Hi Paul,

    > 1. native, non-bidi ... the well-tested code that everyone uses now
    > 2. bidi ... some testing, not enough use 3. Pango-based ... to be
    > written and/or ported as needed

    OK, I see I need to explain the relationship between #1, #2 and #3.

    #2, the bidi code, is not some monster genetically unrelated to #1,
    the non-bidi code. #2 does everything that #1 does, plus the extra
    bidi processing. In fact most of the time the bidi code is basically a
    carbon copy of the non-bidi code just with some extra logic
    plugged into it, often just an extra if(LTR) {} else {}. For left-to-right
    languages it behaves identically to #1, because the code is virtually
    the same, i.e., the if(LTR) branch is usually just a copy of the non-
    bidi code in the #ifndef BIDI_ENABLED #endif section.

    #3, the Pango stuff, is not a replacement for #2, it is replacement
    for the home grown shaping engine that resides withing #2. The
    actual bidi layout code will stay in place for the time being for
    reasons we have discussed at length in the Pango thread.

    > Most of the developers we hope to attract will only care about those
    > languages, and we *definitely* want them working on HEAD. Yet it
    > sounds like you're talking about *removing* that proven code
    > immediately. Why?

    Because the bidi code is a superset of the proven non-bidi code,
    and the whole scenario is a maintanence nightmare. I would be
    surprised if there would not be some extra bugs as a result of this,
    but I doubt that they will be too many. I personally do not use any
    other version of AW than the bidi one and I have not come across
    any bidi-only problems since the start of the year. If you look at the
    bugzilla, you will see that there are only a few bugs reported for the
    bidi build (not necessarily because no-one tests it, there is now a
    number of deticated testers), and AFAIK all of them are related to
    RTL languages.

    > While it may make life easier for Pango development to force all
    > developers to live through that transition as it happens, it makes
    > *much* more sense (for other developers) to *not* switch everyone over
    > until Pango is further along.

    Just to make sure we understand each other, I am not talking
    about switching everyone to Pango, you are quite right, we are
    nowhere near that (we have hardly started). I am talking about
    switching everone to a somewhat expanded version of the proven
    existing code. This change will not break the tree, nor will it reduce
    the functionality for the languages supported by the non-bidi build
    (in fact it will enhance it as the bidi build is more Unicode
    compliant).

    I hope this explain things.

    Tomas



    This archive was generated by hypermail 2.1.4 : Thu May 09 2002 - 05:21:52 EDT