This document describes how libicuuc and libicudata were packaged for build2.
In particular, this understanding will be useful when upgrading to a new
upstream version. See ../README-DEV for general notes on ICU packaging.

Symlink the required upstream directories into libicu/:

$ ln -s ../../upstream/icu4c/source/common libicu/uc

Compiling against the latest C++ standard ends up with the "invalid conversion
from ‘const char8_t*’ to ‘const char*’" errors for libicu/uc/ucasemap.cpp and
so we patch it:

$ cp libicu/uc/ucasemap.cpp libicu
$ patch -p0 <libicu/ucasemap.cpp.patch

Also we fix a security issue CVE-2020-10531 reported by Debian using the
upstream's fix (commit b7d08bc04a4296982fcef8b6b8a354a9e4e7afca) as a base
(dropping the test changes):

$ cp libicu/uc/unistr.cpp libicu
$ patch -p0 <libicu/unistr.cpp.patch

Also we fix the 'unable to find numeric literal operator' MinGW GCC 11 error:

$ cp libicu/uc/putil.cpp libicu
$ patch -p0 <libicu/putil.cpp.patch

And we also fix Clang 10 compilation error:

$ cp libicu/uc/uniset.cpp libicu
$ patch -p0 <libicu/uniset.cpp.patch

Copy the required upstream auto-generated libicudata source files into
libicu/data/, replacing '65' with a major version. Here we assume that the
upstream package is built on the ASCII-based platform with the little-endian
byte ordering. Note that while copying we join the source files by the data
type to speedup compilation.

$ cd libicu/data/
$ rm -r -f le
$ mkdir le
$ cp <upstream-out-dir>/data/out/tmp/icudt65l_dat.c le
$ for f in $(find <upstream-out-dir>/data/out/tmp/ -name '*.c' ! -name icudt65l_dat.c); do
    n="$(sed -n -r -e 's%^(coll|curr|brkitr|lang|locales|rbnf|region|unit|zone)_.*$%\1%p' <<< $(basename "$f"))"
    if [ -z "$n" ]; then
      n="other"
    fi
    cat "$f" >>le/$n.c
  done

Note that we also need to generate libicudata source files for the big-endian
byte ordering, transcoding the auto-generated binary data files and converting
them to C files afterwards:

$ cd <upstream-out-dir>/data

$ export PATH=<upstream-out-dir>/bin:$PATH
$ export LD_LIBRARY_PATH=<upstream-out-dir>/lib:$LD_LIBRARY_PATH

# Transcode.
#
$ for f in $(cd out/build/icudt65l && find -type f); do
    r="out/build/icudt65b/$f"
    mkdir -p "$(dirname "$r")"
    icupkg -tb "out/build/icudt65l/$f" "$r"
  done

# Convert to C files.
#
$ mkdir out/tmp2
$ pkgdata -O icupkg.inc --without-assembly -q -c -s out/build/icudt65b \
-d out/tmp2 -e icudt65 -T out/tmp2 -p icudt65b -m dll -r 65.1 -L icudata \
out/tmp/icudata.lst

Now copy the big-endian data files into libicu/data/, similar to the above:

$ cd libicu/data/
$ rm -r -f be
$ mkdir be
$ cp <upstream-out-dir>/data/out/tmp2/icudt65b_dat.c be
$ for f in $(find <upstream-out-dir>/data/out/tmp2/ -name '*.c' ! -name icudt65b_dat.c); do
    n="$(sed -n -r -e 's%^(coll|curr|brkitr|lang|locales|rbnf|region|unit|zone)_.*$%\1%p' <<< $(basename "$f"))"
    if [ -z "$n" ]; then
      n="other"
    fi
    cat "$f" >>be/$n.c
  done

Note that currently we don't support EBCDIC charset-based platforms that are
not very common these days. Thought, we can support them if requested (see the
-te option description in `icupkg --help` for details).

Manually create libicu/data/config.h and libicu/data/*.c files including the
corresponding type-based data files (le/coll.c, etc) depending on the target
platform byte ordering.