summaryrefslogtreecommitdiff
path: root/macros/latex-dev/required/l3kernel/l3unicode.dtx
diff options
context:
space:
mode:
Diffstat (limited to 'macros/latex-dev/required/l3kernel/l3unicode.dtx')
-rw-r--r--macros/latex-dev/required/l3kernel/l3unicode.dtx1031
1 files changed, 1031 insertions, 0 deletions
diff --git a/macros/latex-dev/required/l3kernel/l3unicode.dtx b/macros/latex-dev/required/l3kernel/l3unicode.dtx
new file mode 100644
index 0000000000..0ee6858f5b
--- /dev/null
+++ b/macros/latex-dev/required/l3kernel/l3unicode.dtx
@@ -0,0 +1,1031 @@
+% \iffalse meta-comment
+%
+%% File: l3unicode.dtx
+%
+% Copyright (C) 2018-2024 The LaTeX Project
+%
+% It may be distributed and/or modified under the conditions of the
+% LaTeX Project Public License (LPPL), either version 1.3c of this
+% license or (at your option) any later version. The latest version
+% of this license is in the file
+%
+% https://www.latex-project.org/lppl.txt
+%
+% This file is part of the "l3kernel bundle" (The Work in LPPL)
+% and all files in that bundle must be distributed together.
+%
+% -----------------------------------------------------------------------
+%
+% The development version of the bundle can be found at
+%
+% https://github.com/latex3/latex3
+%
+% for those people who are interested.
+%
+%<*driver>
+\documentclass[full,kernel]{l3doc}
+\begin{document}
+ \DocInput{\jobname.dtx}
+\end{document}
+%</driver>
+% \fi
+%
+% \title{^^A
+% The \pkg{l3unicode} module\\ Unicode support functions^^A
+% }
+%
+% \author{^^A
+% The \LaTeX{} Project\thanks
+% {^^A
+% E-mail:
+% \href{mailto:latex-team@latex-project.org}
+% {latex-team@latex-project.org}^^A
+% }^^A
+% }
+%
+% \date{Released 2024-04-11}
+%
+% \maketitle
+%
+% \begin{documentation}
+%
+% This module provides Unicode-specific functions along with loading data
+% from a range of Unicode Consortium files. Most of the code here is
+% internal, but there are a small set of public functions. These work with
+% Unicode \meta{codepoints} and are designed to give useable results with
+% both Unicode-aware and $8$-bit engines.
+%
+% \begin{function}[EXP, added = 2022-10-09, updated = 2022-11-09]
+% {\codepoint_generate:nn}
+% \begin{syntax}
+% \cs{codepoint_generate:nn} \Arg{codepoint} \Arg{catcode}
+% \end{syntax}
+% Generates one or more character tokens representing the \meta{codepoint}.
+% With Unicode engines, exactly one character token will be generated, and
+% this will have the \meta{catcode} specified as the second argument:
+% \begin{itemize}
+% \item $1$ (begin group)
+% \item $2$ (end group)
+% \item $3$ (math toggle)
+% \item $4$ (alignment)
+% \item $6$ (parameter)
+% \item $7$ (math superscript)
+% \item $8$ (math subscript)
+% \item $10$ (space)
+% \item $11$ (letter)
+% \item $12$ (other)
+% \item $13$ (active)
+% \end{itemize}
+% For $8$-bit engines, between one and four character tokens will be
+% produced: these will be the bytes of the UTF-8 representation of the
+% \meta{codepoint}. For all codepoints outside of the classical ASCII
+% range, the generated character tokens will be active (category code
+% $13$); for codepoints in the ASCII range, the given \meta{catcode}
+% will be used. To allow the result of this function to be used
+% inside an expansion context, the result is protected by \cs{exp_not:n}.
+%
+% \begin{texnote}
+% Users of (u)p\TeX{} note that these engines are treated as $8$-bit in
+% this context. In particular, for up\TeX{}, irrespective of the
+% \tn{kcatcode} of the \meta{codepoint}, any value outside the ASCII range
+% will result in a series of active bytes being generated.
+% \end{texnote}
+% \end{function}
+%
+% \begin{function}[EXP, added = 2022-10-09]
+% {\codepoint_str_generate:n}
+% \begin{syntax}
+% \cs{codepoint_str_generate:n} \Arg{codepoint}
+% \end{syntax}
+% Generates one or more character tokens representing the \meta{codepoint}.
+% With Unicode engines, exactly one character token will be generated.
+% For $8$-bit engines, between one and four character tokens will be
+% produced: these will be the bytes of the UTF-8 representation of the
+% \meta{codepoint}. All of the generated character tokens will be of
+% category code $12$, except any spaces (codepoint $32$), which will be
+% category code $10$.
+% \end{function}
+%
+% \begin{function}[added = 2023-06-19, EXP]{\codepoint_to_category:n}
+% \begin{syntax}
+% \cs{codepoint_to_category:n} \Arg{codepoint}
+% \end{syntax}
+% Expands to the Unicode general category identifier of the \meta{codepoint}.
+% The general category identifier is a string made up of two letter
+% characters, the first uppercase and the second lowercase. The uppercase
+% letters divide codepoints into broader groups, which are then refined
+% by the lowercase letter. For example, codepoints representing letters
+% all have identifiers starting \texttt{L}, for example \texttt{Lu}
+% (uppercase letter), \texttt{Lt} (titlecase letter), \emph{etc.}
+% Full details are available in the documentation provided by the Unicode
+% Consortium: see
+% \url{https://www.unicode.org/reports/tr44/#General_Category_Values}
+% \end{function}
+%
+% \begin{function}[added = 2022-10-09, EXP]{\codepoint_to_nfd:n}
+% \begin{syntax}
+% \cs{codepoint_to_nfd:n} \Arg{codepoint}
+% \end{syntax}
+% Converts the \meta{codepoint} to the Unicode Normalization
+% Form Canonical Decomposition. The generated character(s) will have
+% the current category code as they would if typed in directly for Unicode
+% engines; for $8$-bit engines, active characters are used for all codepoints
+% outside of the ASCII range.
+% \end{function}
+%
+% \end{documentation}
+%
+% \begin{implementation}
+%
+% \section{\pkg{l3unicode} implementation}
+%
+% \begin{macrocode}
+%<*package>
+% \end{macrocode}
+%
+% \begin{macrocode}
+%<@@=codepoint>
+% \end{macrocode}
+%
+% \subsection{User functions}
+%
+% \begin{macro}[EXP]{\codepoint_str_generate:n}
+% \begin{macro}[EXP]{\@@_str_generate:nnnn}
+% \begin{macro}[EXP]{\codepoint_generate:nn}
+% \begin{macro}[EXP]{\@@_generate:nnnn}
+% \begin{macro}[EXP]{\@@_generate:n}
+% Conversion of a codepoint to a character (Unicode engines) or to one
+% or more bytes ($8$-bit engines) is required. For loading the data,
+% all that is needed is the form which creates strings: these are outside
+% the group as they will also be used when looking up data in the hash
+% table storage at point-of-use. Later, we will also need functions that
+% can generate character tokens for document use: those are defined below,
+% in the data recovery setup.
+% \begin{macrocode}
+\bool_lazy_or:nnTF
+ { \sys_if_engine_luatex_p: }
+ { \sys_if_engine_xetex_p: }
+ {
+ \cs_new:Npn \codepoint_str_generate:n #1
+ {
+ \int_compare:nNnTF {#1} = { `\ }
+ { ~ }
+ { \char_generate:nn {#1} { 12 } }
+ }
+ \cs_new:Npn \codepoint_generate:nn #1#2
+ {
+ \int_compare:nNnTF {#1} = { `\ }
+ { ~ }
+ {
+ \__kernel_exp_not:w \exp_after:wN \exp_after:wN \exp_after:wN
+ { \char_generate:nn {#1} {#2} }
+ }
+ }
+ }
+ {
+ \cs_new:Npn \codepoint_str_generate:n #1
+ {
+ \int_compare:nNnTF {#1} = { `\ }
+ { ~ }
+ {
+ \use:e
+ {
+ \exp_not:N \@@_str_generate:nnnn
+ \__kernel_codepoint_to_bytes:n {#1}
+ }
+ }
+ }
+ \cs_new:Npn \@@_str_generate:nnnn #1#2#3#4
+ {
+ \char_generate:nn {#1} { 12 }
+ \tl_if_blank:nF {#2}
+ {
+ \char_generate:nn {#2} { 12 }
+ \tl_if_blank:nF {#3}
+ {
+ \char_generate:nn {#3} { 12 }
+ \tl_if_blank:nF {#4}
+ { \char_generate:nn {#4} { 12 } }
+ }
+ }
+ }
+ \cs_new:Npn \codepoint_generate:nn #1#2
+ {
+ \int_compare:nNnTF {#1} = { `\ }
+ { ~ }
+ {
+ \int_compare:nNnTF {#1} < { "80 }
+ {
+ \__kernel_exp_not:w \exp_after:wN \exp_after:wN \exp_after:wN
+ { \char_generate:nn {#1} {#2} }
+ }
+ {
+ \use:e
+ {
+ \exp_not:N \@@_generate:nnnn
+ \__kernel_codepoint_to_bytes:n {#1}
+ }
+ }
+ }
+ }
+ \cs_new:Npn \@@_generate:nnnn #1#2#3#4
+ {
+ \__kernel_exp_not:w \exp_after:wN
+ {
+ \tex_expanded:D
+ {
+ \@@_generate:n {#1}
+ \@@_generate:n {#2}
+ \tl_if_blank:nF {#3}
+ {
+ \@@_generate:n {#3}
+ \tl_if_blank:nF {#4}
+ { \@@_generate:n {#4} }
+ }
+ }
+ }
+ }
+ \cs_new:Npn \@@_generate:n #1
+ {
+ \__kernel_exp_not:w \exp_after:wN \exp_after:wN \exp_after:wN
+ { \char_generate:nn {#1} { 13 } }
+ }
+ }
+% \end{macrocode}
+% \end{macro}
+% \end{macro}
+% \end{macro}
+% \end{macro}
+% \end{macro}
+%
+% \begin{macro}[EXP]{\__kernel_codepoint_to_bytes:n}
+% \begin{macro}[EXP]{\@@_to_bytes_auxi:n}
+% \begin{macro}[EXP]{\@@_to_bytes_auxii:Nnn}
+% \begin{macro}[EXP]{\@@_to_bytes_auxiii:n}
+% \begin{macro}[EXP]
+% {
+% \@@_to_bytes_outputi:nw ,
+% \@@_to_bytes_outputii:nw ,
+% \@@_to_bytes_outputiii:nw ,
+% \@@_to_bytes_outputiv:nw
+% }
+% \begin{macro}[EXP]
+% {\@@_to_bytes_output:nnn, \@@_to_bytes_output:fnn}
+% \begin{macro}[EXP]{\@@_to_bytes_end:}
+% This code converts a codepoint into the correct UTF-8 representation.
+% In terms of the algorithm itself, see
+% \url{https://en.wikipedia.org/wiki/UTF-8} for the octet pattern.
+% \begin{macrocode}
+\cs_new:Npn \__kernel_codepoint_to_bytes:n #1
+ {
+ \exp_args:Nf \@@_to_bytes_auxi:n
+ { \int_eval:n {#1} }
+ }
+\cs_new:Npn \@@_to_bytes_auxi:n #1
+ {
+ \if_int_compare:w #1 > "80 \exp_stop_f:
+ \if_int_compare:w #1 < "800 \exp_stop_f:
+ \@@_to_bytes_outputi:nw
+ { \@@_to_bytes_auxii:Nnn C {#1} { 64 } }
+ \@@_to_bytes_outputii:nw
+ { \@@_to_bytes_auxiii:n {#1} }
+ \else:
+ \if_int_compare:w #1 < "10000 \exp_stop_f:
+ \@@_to_bytes_outputi:nw
+ { \@@_to_bytes_auxii:Nnn E {#1} { 64 * 64 } }
+ \@@_to_bytes_outputii:nw
+ {
+ \@@_to_bytes_auxiii:n
+ { \int_div_truncate:nn {#1} { 64 } }
+ }
+ \@@_to_bytes_outputiii:nw
+ { \@@_to_bytes_auxiii:n {#1} }
+ \else:
+ \@@_to_bytes_outputi:nw
+ {
+ \@@_to_bytes_auxii:Nnn F
+ {#1} { 64 * 64 * 64 }
+ }
+ \@@_to_bytes_outputii:nw
+ {
+ \@@_to_bytes_auxiii:n
+ { \int_div_truncate:nn {#1} { 64 * 64 } }
+ }
+ \@@_to_bytes_outputiii:nw
+ {
+ \@@_to_bytes_auxiii:n
+ { \int_div_truncate:nn {#1} { 64 } }
+ }
+ \@@_to_bytes_outputiv:nw
+ { \@@_to_bytes_auxiii:n {#1} }
+ \fi:
+ \fi:
+ \else:
+ \@@_to_bytes_outputi:nw {#1}
+ \fi:
+ \@@_to_bytes_end: { } { } { } { }
+ }
+\cs_new:Npn \@@_to_bytes_auxii:Nnn #1#2#3
+ { "#10 + \int_div_truncate:nn {#2} {#3} }
+\cs_new:Npn \@@_to_bytes_auxiii:n #1
+ { \int_mod:nn {#1} { 64 } + 128 }
+\cs_new:Npn \@@_to_bytes_outputi:nw
+ #1 #2 \@@_to_bytes_end: #3
+ { \@@_to_bytes_output:fnn { \int_eval:n {#1} } { } {#2} }
+\cs_new:Npn \@@_to_bytes_outputii:nw
+ #1 #2 \@@_to_bytes_end: #3#4
+ { \@@_to_bytes_output:fnn { \int_eval:n {#1} } { {#3} } {#2} }
+\cs_new:Npn \@@_to_bytes_outputiii:nw
+ #1 #2 \@@_to_bytes_end: #3#4#5
+ {
+ \@@_to_bytes_output:fnn
+ { \int_eval:n {#1} } { {#3} {#4} } {#2}
+ }
+\cs_new:Npn \@@_to_bytes_outputiv:nw
+ #1 #2 \@@_to_bytes_end: #3#4#5#6
+ {
+ \@@_to_bytes_output:fnn
+ { \int_eval:n {#1} } { {#3} {#4} {#5} } {#2}
+ }
+\cs_new:Npn \@@_to_bytes_output:nnn #1#2#3
+ {
+ #3
+ \@@_to_bytes_end: #2 {#1}
+ }
+\cs_generate_variant:Nn \@@_to_bytes_output:nnn { f }
+\cs_new:Npn \@@_to_bytes_end: { }
+% \end{macrocode}
+% \end{macro}
+% \end{macro}
+% \end{macro}
+% \end{macro}
+% \end{macro}
+% \end{macro}
+% \end{macro}
+%
+% \begin{macro}[EXP]{\codepoint_to_category:n}
+% Get the value and convert back to the string.
+% \begin{macrocode}
+\cs_new:Npn \codepoint_to_category:n #1
+ {
+ \cs:w
+ c_@@_category_
+ \tex_romannumeral:D
+ \__kernel_codepoint_data:nn { category } {#1}
+ _str
+ \cs_end:
+ }
+% \end{macrocode}
+% \end{macro}
+%
+% \begin{macro}[EXP]{\codepoint_to_nfd:n, \@@_to_nfd:n}
+% \begin{macro}[EXP]{\@@_to_nfd:nn}
+% \begin{macro}[EXP]{\@@_to_nfd:nnn}
+% \begin{macro}[EXP]{\@@_to_nfd:nnnn}
+% Converted to NFD is a potentially-recursive process: the key is to
+% check if we get the input codepoint back again. As far as possible,
+% we use the same path for all engines.
+% \begin{macrocode}
+\cs_new:Npn \codepoint_to_nfd:n #1
+ { \exp_args:Ne \@@_to_nfd:n { \int_eval:n {#1} } }
+\cs_new:Npn \@@_to_nfd:n #1
+ { \@@_to_nfd:nn {#1} { \char_value_catcode:n {#1} } }
+\bool_lazy_or:nnF
+ { \sys_if_engine_luatex_p: }
+ { \sys_if_engine_xetex_p: }
+ {
+ \cs_gset:Npn \@@_to_nfd:n #1
+ {
+ \int_compare:nNnTF {#1} > { "80 }
+ { \@@_to_nfd:nn {#1} { 12 } }
+ { \@@_to_nfd:nn {#1} { \char_value_catcode:n {#1} } }
+ }
+ }
+\cs_new:Npn \@@_to_nfd:nn #1#2
+ {
+ \exp_args:Ne \@@_to_nfd:nnn
+ { \@@_nfd:n {#1} } {#1} {#2}
+ }
+\cs_new:Npn \@@_to_nfd:nnn #1#2#3 { \@@_to_nfd:nnnn #1 {#2} {#3} }
+\cs_new:Npn \@@_to_nfd:nnnn #1#2#3#4
+ {
+ \int_compare:nNnTF {#1} = {#3}
+ { \codepoint_generate:nn {#1} {#4} }
+ {
+ \@@_to_nfd:nn {#1} {#4}
+ \tl_if_blank:nF {#2}
+ { \@@_to_nfd:nn {#2} {#4} }
+ }
+ }
+% \end{macrocode}
+% \end{macro}
+% \end{macro}
+% \end{macro}
+% \end{macro}
+%
+% \subsection{Data loader}
+%
+% Text operations requires data from the Unicode Consortium. Data read into
+% Unicode engine formats is at best a small part of what we need, so there
+% is a loader here to set up the appropriate data structures.
+%
+% Where we need data for most or all of the Unicode range, we use the two-stage
+% table approach recommended by the Unicode Consortium and demonstrated in a
+% model implementation in Python in
+% \url{https://www.strchr.com/multi-stage_tables}. This approach uses the
+% \texttt{intarray} (\texttt{fontdimen}-based) data type as it is fast for
+% random access and avoids significant hash table usage. In contrast, where
+% only a small subset of codepoints are required, storage as macros is
+% preferable. There is also some consideration of the effort needed to load
+% data: see for example the grapheme breaking information, which would be
+% problematic to convert into a two-stage table but which can be used with
+% reasonable performance in a small number of comma lists (at the cost that
+% breaking at higher codepoint Hangul characters will be slightly slow).
+%
+% \begin{variable}{\c_@@_block_size_int}
+% Choosing the block size for the blocks in the two-stage approach is
+% non-trivial: depending on the data stored, the optimal size for
+% memory usage will vary. At the same time, for us there is also the
+% question of load-time: larger blocks require longer comma lists
+% as intermediates, so are slower. As this is going to be needed
+% to use the data, we set it up outside of the group for clarity.
+% \begin{macrocode}
+\int_const:Nn \c_@@_block_size_int { 64 }
+% \end{macrocode}
+% \end{variable}
+%
+% Parsing the data files can be the same way for all engines, but where they
+% are stored as character tokens, the construction method depends on whether
+% they are Unicode or $8$-bit internally. Parsing is therefore done by common
+% functions, with some data storage using engine-specific auxiliaries.
+%
+% As only the data needs to remain at the end of this process, everything
+% is set up inside a group. The only thing that is outside is creating a
+% stream: they are global anyway and it is best to force a stream for
+% all engines.
+%
+% \begin{variable}{\g_@@_data_ior}
+% \begin{macrocode}
+\ior_new:N \g_@@_data_ior
+% \end{macrocode}
+% \end{variable}
+%
+% We need some setup for the two-part table approach. The number of blocks we
+% need will be variable, but the resulting size of the stage one table
+% is predictable. For performance reasons, we therefore create the stage one
+% tables now so they can be used immediately, and will later rename them as a
+% constant tables. For each two-stage table construction, we need a comma
+% list to hold the partial block and a couple of integers to track where
+% we are up to. To avoid burning registers, the latter are stored in macros
+% and are \enquote{fake} integers. We also avoid any \texttt{new} functions,
+% keeping as much as possible local.
+%
+% As we need both positive and negative values, case data requires one
+% two-stage table for each transformation. In contrasts, general Unicode
+% properties could be stored in one table with appropriate combination rules:
+% that is not done at present but is likely to be added over time. Here, all
+% that is needed is additional entries into the comma-list to create the
+% structures.
+%
+% Notice that in the standard \pkg{expl3} way we are indexes position not
+% offset: that does mean a little work later.
+% \begin{macrocode}
+\group_begin:
+ \clist_map_inline:nn
+ { category , uppercase , lowercase }
+ {
+ \cs_set_nopar:cpn { l_@@_ #1 _block_clist } { }
+ \cs_set_nopar:cpn { l_@@_ #1 _block_tl } { 1 }
+ \cs_set_nopar:cpn { l_@@_ #1 _pos_tl } { 0 }
+ \intarray_new:cn { g_@@_ #1 _index_intarray }
+ { \int_div_truncate:nn { "110000 } \c_@@_block_size_int }
+ }
+% \end{macrocode}
+% We need an integer value when matching the current block to those we have
+% already seen, and a way to track codepoints for handling ranges. Again,
+% we avoid using up registers or creating global names.
+% \begin{macrocode}
+ \cs_set_nopar:Npn \l_@@_next_codepoint_fint_tl { 0 }
+ \cs_set_nopar:Npn \l_@@_matched_block_tl { 0 }
+% \end{macrocode}
+% For Unicode general category, there needs to be numerical representation of
+% each possible value. As we need to go from string to number here, but the
+% other way elsewhere, we set up fast mappings both ways, but one set local
+% and the other as constants.
+% \begin{macrocode}
+ \cs_set_protected:Npn \@@_data_auxi:w #1#2
+ {
+ \quark_if_recursion_tail_stop:n {#2}
+ \cs_set_nopar:cpn { l_@@_category_ #2 _tl } {#1}
+ \str_const:cn { c_@@_category_ \tex_romannumeral:D #1 _str } {#2}
+ \exp_args:Ne \@@_data_auxi:w { \int_eval:n { #1 + 1 } }
+ }
+ \@@_data_auxi:w { 1 }
+ { Lu } { Ll } { Lt } { Lm } { Lo }
+ { Mn } { Me } { Mc }
+ { Nd } { Nl } { No }
+ { Zs } { Zl } { Zp }
+ { Cc } { Cf } { Co } { Cs } { Cn }
+ { Pd } { Ps } { Pe } { Pc } { Po } { Pi } { Pf }
+ { Sm } { Sc } { Sk } { So }
+ \q_recursion_tail
+ \q_recursion_stop
+% \end{macrocode}
+% Parse the main Unicode data file and pull out the NFD and case changing
+% data. The NFD data is stored on using the hash table approach and can yield
+% a predictable number of codepoints: one or two. We also need the case data,
+% which will be modified further below. To allow for finding ranges, the
+% description of the codepoint needs to be carried forward.
+% \begin{macrocode}
+ \cs_set_protected:Npn \@@_data_auxi:w
+ #1 ; #2 ; #3 ; #4 ; #5 ; #6 ; #7 ; #8 ; #9 ;
+ {
+ \tl_if_blank:nF {#6}
+ {
+ \tl_if_head_eq_charcode:nNF {#6} < % >
+ { \@@_data_auxii:w #1 ; #6 ~ \q_stop }
+ }
+ \@@_data_auxiii:w #1 ; #2 ; #3 ;
+ }
+ \cs_set_protected:Npn \@@_data_auxii:w #1 ; #2 ~ #3 \q_stop
+ {
+ \tl_const:ce
+ { c_@@_nfd_ \codepoint_str_generate:n {"#1} _tl }
+ {
+ {"#2}
+ { \tl_if_blank:nF {#3} {"#3} }
+ }
+ }
+% \end{macrocode}
+% The category data needs to be converted from a string to the numerical
+% equivalent: a simple operation.
+% The case data is going to be stored as an offset from the parent character,
+% rather than an absolute value. We therefore deal with that plus the situation
+% where a codepoint has no mapping data in one shot.
+% \begin{macrocode}
+ \cs_set_protected:Npn \@@_data_auxiii:w
+ #1 ; #2 ; #3 ; #4 ; #5 ; #6 ; #7 ; #8 ; #9 ~ \q_stop
+ {
+ \use:e
+ {
+ \@@_data_auxiv:w
+ #1 ; #2 ;
+ \@@_data_category:n {#3} ;
+ \@@_data_offset:nn {#1} {#7} ;
+ \@@_data_offset:nn {#1} {#8} ;
+ #9;
+ }
+ }
+ \cs_set:Npn \@@_data_category:n #1
+ { \use:c { l_@@_category_ #1 _tl } }
+ \cs_set:Npn \@@_data_offset:nn #1#2
+ {
+ \tl_if_blank:nTF {#2}
+ { 0 }
+ { \int_eval:n { "#2 - "#1 } }
+ }
+% \end{macrocode}
+% To deal with ranges, we track the position of the next codepoint expected.
+% If there is a gap, we deal with that separately: it could be a range or
+% an unused part of the Unicode space. As such, we deal with the current
+% codepoint here whether or not there is range to fill in. Upper- and
+% lowercase data go into the two-stage table, any titlecase exception is
+% just stored in a macro. The data for the codepoint is added to the current
+% block, and if that is now complete we move on to save the block. The
+% case exceptions are all stored as codepoints, with a fixed number of
+% balanced text as we know that there are never more than three.
+% \begin{macrocode}
+ \cs_set_protected:Npn \@@_data_auxiv:w #1 ; #2 ; #3 ; #4 ; #5 ; #6 ;
+ {
+ \int_compare:nNnT {"#1} > \l_@@_next_codepoint_fint_tl
+ {
+ \@@_data_auxv:nnnnw {#1} {#3} {#4} {#5}
+ #2 Last> \q_stop
+ }
+ \@@_add:nn { category } {#3}
+ \@@_add:nn { uppercase } {#4}
+ \@@_add:nn { lowercase } {#5}
+ \int_compare:nNnF {#4} = { \@@_data_offset:nn {#1} {#6} }
+ {
+ \tl_const:ce
+ { c_@@_titlecase_ \codepoint_str_generate:n {"#1} _tl }
+ { {"#6} { } { } }
+ }
+ \tl_set:Ne \l_@@_next_codepoint_fint_tl
+ { \int_eval:n { "#1 + 1 } }
+ }
+ \cs_set_protected:Npn \@@_add:nn #1#2
+ {
+ \clist_put_right:cn { l_@@_ #1 _block_clist } {#2}
+ \int_compare:nNnT { \clist_count:c { l_@@_ #1 _block_clist } }
+ = \c_@@_block_size_int
+ { \@@_save_blocks:nn {#1} { 1 } }
+ }
+% \end{macrocode}
+% Distinguish between a range and a gap, and pass on the appropriate value(s).
+% The general category for unassigned characters is \texttt{Cn}, so we
+% find the correct value once and then use that.
+% \begin{macrocode}
+ \cs_set_protected:Npe \@@_data_auxv:nnnnw #1#2#3#4#5 Last> #6 \q_stop
+ {
+ \exp_not:N \tl_if_blank:nTF {#6}
+ {
+ \exp_not:N \@@_range:nnn {#1} { category }
+ { \exp_not:V \l_@@_category_Cn_tl }
+ \exp_not:N \@@_range:nnn {#1} { uppercase } { 0 }
+ \exp_not:N \@@_range:nnn {#1} { lowercase } { 0 }
+ }
+ {
+ \exp_not:N \@@_range:nnn {#1} { category } {#2}
+ \exp_not:N \@@_range:nnn {#1} { uppercase } {#3}
+ \exp_not:N \@@_range:nnn {#1} { lowercase } {#4}
+ }
+ }
+% \end{macrocode}
+% Calculated the length of the range and the space remaining in the current
+% block.
+% \begin{macrocode}
+ \cs_set_protected:Npn \@@_range:nnn #1
+ {
+ \exp_args:Nf \@@_range_aux:nnn
+ { \int_eval:n { "#1 - \l_@@_next_codepoint_fint_tl } }
+ }
+ \cs_set_protected:Npn \@@_range_aux:nnn #1#2
+ {
+ \exp_args:Nf \@@_range:nnnn
+ {
+ \int_min:nn
+ {#1}
+ {
+ \c_@@_block_size_int
+ - \clist_count:c { l_@@_ #2 _block_clist }
+ }
+ }
+ {#1} {#2}
+ }
+% \end{macrocode}
+% Here we want to do three things: add to and possibly complete the current
+% block, add complete blocks quickly, then finish up the range in a final
+% open block. We need to avoid as far as possible avoid dealing with every
+% single codepoint, so the middle step is optimised.
+% \begin{macrocode}
+ \cs_set_protected:Npn \@@_range:nnnn #1#2#3#4
+ {
+ \prg_replicate:nn {#1}
+ { \clist_put_right:cn { l_@@_ #3 _block_clist } {#4} }
+ \int_compare:nNnT { \clist_count:c { l_@@_ #3 _block_clist } }
+ = \c_@@_block_size_int
+ { \@@_save_blocks:nn {#3} { 1 } }
+ \int_compare:nNnF
+ { \int_div_truncate:nn { #2 - #1 } \c_@@_block_size_int } = 0
+ {
+ \tl_set:ce { l_@@_ #3 _block_clist }
+ {
+ \exp_args:NNe \use:nn \use_none:n
+ { \prg_replicate:nn { \c_@@_block_size_int } { , #4 } }
+ }
+ \@@_save_blocks:nn {#3}
+ { \int_div_truncate:nn { (#2 - #1) } \c_@@_block_size_int }
+ }
+ \prg_replicate:nn
+ { \int_mod:nn { #2 - #1 } \c_@@_block_size_int }
+ { \clist_put_right:ce { l_@@_ #3 _block_clist } {#4} }
+ }
+% \end{macrocode}
+% To allow rapid comparison, each completed block is stored locally as a
+% comma list: once all of the blocks have been created, they are converted
+% into an \texttt{intarray} in one step. The aim here is to check the current
+% block against those we've already used, and either match to an existing
+% block or save a new block.
+% \begin{macrocode}
+ \cs_set_protected:Npn \@@_save_blocks:nn #1#2
+ {
+ \tl_set_eq:Nc \l_@@_matched_block_tl { l_@@_ #1 _block_tl }
+ \int_step_inline:nn { \tl_use:c { l_@@_ #1 _block_tl } - 1 }
+ {
+ \tl_if_eq:ccT { l_@@_ #1 _block_clist }
+ { l_@@_ #1 _block_ ##1 _clist }
+ { \tl_set:Nn \l_@@_matched_block_tl {##1} }
+ }
+ \int_compare:nNnT
+ { \tl_use:c { l_@@_ #1 _block_tl } } = \l_@@_matched_block_tl
+ {
+ \clist_set_eq:cc
+ {
+ l_@@_ #1 _block_
+ \tl_use:c { l_@@_ #1 _block_tl } _clist
+ }
+ { l_@@_ #1 _block_clist }
+ \tl_set:ce { l_@@_ #1 _block_tl }
+ { \int_eval:n { \tl_use:c { l_@@_ #1 _block_tl } + 1 } }
+ }
+ \prg_replicate:nn {#2}
+ {
+ \tl_set:ce { l_@@_ #1 _pos_tl }
+ { \int_eval:n { \tl_use:c { l_@@_ #1 _pos_tl } + 1 } }
+ \exp_args:Nc \__kernel_intarray_gset:Nnn
+ { g_@@_ #1 _index_intarray }
+ { \tl_use:c { l_@@_ #1 _pos_tl } }
+ \l_@@_matched_block_tl
+ }
+ \clist_clear:c { l_@@_ #1 _block_clist }
+ }
+% \end{macrocode}
+% Close out the final block, rename the first stage table, then combine all
+% of the block comma-lists into one large second-stage table with offsets.
+% As we use an index not an offset, there is a little back-and-forward to do.
+% \begin{macrocode}
+ \cs_set_protected:Npn \@@_finalise_blocks:
+ {
+ \clist_map_inline:nn { category , uppercase , lowercase }
+ {
+ \@@_range:nnn { 110000 } {##1} { 0 }
+ \@@_finalise_blocks:n {##1}
+ }
+ }
+ \cs_set_protected:Npn \@@_finalise_blocks:n #1
+ {
+ \cs_gset_eq:cc { c_@@_ #1 _index_intarray } { g_@@_ #1 _index_intarray }
+ \cs_undefine:c { g_@@_ #1 _index_intarray }
+ \intarray_new:cn { g_@@_ #1 _blocks_intarray }
+ { ( \tl_use:c { l_@@_ #1 _block_tl } - 1 ) * \c_@@_block_size_int }
+ \int_step_inline:nn { \tl_use:c { l_@@_ #1 _block_tl } - 1 }
+ {
+ \exp_args:Nv \@@_finalise_blocks:nnn
+ { l_@@_ #1 _block_ ##1 _clist }
+ {##1} {#1}
+ }
+ \cs_gset_eq:cc { c_@@_ #1 _blocks_intarray }
+ { g_@@_ #1 _blocks_intarray }
+ \cs_undefine:c { g_@@_ #1 _blocks_intarray }
+ }
+ \cs_set_protected:Npn \@@_finalise_blocks:nnn #1#2#3
+ {
+ \exp_args:Nnf \@@_finalise_blocks:nnnw { 1 }
+ { \int_eval:n { ( #2 - 1 ) * \c_@@_block_size_int } }
+ {#3}
+ #1 , \q_recursion_tail , \q_recursion_stop
+ }
+ \cs_set_protected:Npn \@@_finalise_blocks:nnnw #1#2#3#4 ,
+ {
+ \quark_if_recursion_tail_stop:n {#4}
+ \intarray_gset:cnn { g_@@_ #3 _blocks_intarray }
+ { #1 + #2 }
+ {#4}
+ \exp_args:Nf \@@_finalise_blocks:nnnw
+ { \int_eval:n { #1 + 1 } } {#2} {#3}
+ }
+% \end{macrocode}
+% With the setup done, read the main data file: it's easiest to do that as
+% a token list with spaces retained.
+% \begin{macrocode}
+ \ior_open:Nn \g_@@_data_ior { UnicodeData.txt }
+ \group_begin:
+ \char_set_catcode_space:n { `\ }%
+ \ior_map_variable:NNn \g_@@_data_ior \l_@@_tmpa_tl
+ {%
+ \if_meaning:w \l_@@_tmpa_tl \c_space_tl
+ \exp_after:wN \ior_map_break:
+ \fi:
+ \exp_after:wN \@@_data_auxi:w \l_@@_tmpa_tl \q_stop
+ }%
+ \@@_finalise_blocks:
+ \group_end:
+\group_end:
+% \end{macrocode}
+%
+% \begin{macro}[EXP]{\__kernel_codepoint_data:nn}
+% \begin{macro}[EXP]{\@@_data:nnn}
+% Recover data from a two-stage table: entirely generic as this applies to
+% all tables (as we use the same block size for all of them). Notice that
+% as we use indices not offsets we have to shuffle out-by-one issues. This
+% function is needed \emph{before} loading the special casing data, as there
+% we need to be able to check the standard case mappings.
+% \begin{macrocode}
+\cs_new:Npn \__kernel_codepoint_data:nn #1#2
+ {
+ \exp_args:Nf \@@_data:nnn
+ {
+ \int_eval:n
+ {
+ \c_@@_block_size_int *
+ (
+ \intarray_item:cn { c_@@_ #1 _index_intarray }
+ {
+ \int_div_truncate:nn {#2}
+ \c_@@_block_size_int
+ + 1
+ }
+ - 1
+ )
+ }
+ }
+ {#2} {#1}
+ }
+\cs_new:Npn \@@_data:nnn #1#2#3
+ {
+ \intarray_item:cn { c_@@_ #3 _blocks_intarray }
+ { #1 + \int_mod:nn {#2} \c_@@_block_size_int + 1 }
+ }
+% \end{macrocode}
+% \end{macro}
+% \end{macro}
+%
+% The other data files all use C-style comments so we have to worry about
+% |#| tokens (and reading as strings). The set up for case folding is in two
+% parts. For the basic (core) mappings, folding is the same as lower casing in
+% most positions so only store the differences. For the more complex foldings,
+% always store the result, splitting up the two or three code points in the input
+% as required.
+% \begin{macrocode}
+\group_begin:
+ \ior_open:Nn \g_@@_data_ior { CaseFolding.txt }
+ \cs_set_protected:Npn \@@_data_auxi:w #1 ;~ #2 ;~ #3 ; #4 \q_stop
+ {
+ \if:w \tl_head:n { #2 ? } C
+ \reverse_if:N \if_int_compare:w
+ \int_eval:n { \__kernel_codepoint_data:nn { lowercase } {"#1} + "#1 }
+ = "#3 ~
+ \tl_const:ce
+ { c_@@_casefold_ \codepoint_str_generate:n {"#1} _tl }
+ { {"#3} { } { } }
+ \fi:
+ \else:
+ \if:w \tl_head:n { #2 ? } F
+ \@@_data_auxii:w #1 ~ #3 ~ \q_stop
+ \fi:
+ \fi:
+ }
+% \end{macrocode}
+% Here, |#4| can have a trailing space, so we tidy up a bit at the cost of
+% speed for these small number of cases it applies to.
+% \begin{macrocode}
+ \cs_set_protected:Npn \@@_data_auxii:w #1 ~ #2 ~ #3 ~ #4 \q_stop
+ {
+ \tl_const:ce { c_@@_casefold_ \codepoint_str_generate:n {"#1} _tl }
+ {
+ {"#2}
+ {"#3}
+ { \tl_if_blank:nF {#4} { " \int_to_Hex:n {"#4} } }
+ }
+ }
+ \ior_str_map_inline:Nn \g_@@_data_ior
+ {
+ \reverse_if:N \if:w \c_hash_str \tl_head:w #1 \c_hash_str \q_stop
+ \@@_data_auxi:w #1 \q_stop
+ \fi:
+ }
+ \ior_close:N \g_@@_data_ior
+% \end{macrocode}
+% For upper- and lowercasing special situations, there is a bit more to
+% do as we also have titlecasing to consider, plus we need to stop part-way
+% through the file.
+% \begin{macrocode}
+ \ior_open:Nn \g_@@_data_ior { SpecialCasing.txt }
+ \cs_set_protected:Npn \@@_data_auxi:w
+ #1 ;~ #2 ;~ #3 ;~ #4 ; #5 \q_stop
+ {
+ \use:n { \@@_data_auxii:w #1 ~ lower ~ #2 ~ } ~ \q_stop
+ \use:n { \@@_data_auxii:w #1 ~ upper ~ #4 ~ } ~ \q_stop
+ \str_if_eq:nnF {#3} {#4}
+ { \use:n { \@@_data_auxii:w #1 ~ title ~ #3 ~ } ~ \q_stop }
+ }
+ \cs_set_protected:Npn \@@_data_auxii:w
+ #1 ~ #2 ~ #3 ~ #4 ~ #5 \q_stop
+ {
+ \tl_if_empty:nF {#4}
+ {
+ \tl_const:ce { c_@@_ #2 case_ \codepoint_str_generate:n {"#1} _tl }
+ {
+ {"#3}
+ {"#4}
+ { \tl_if_blank:nF {#5} {"#5} }
+ }
+ }
+ }
+ \ior_str_map_inline:Nn \g_@@_data_ior
+ {
+ \str_if_eq:eeTF { \tl_head:w #1 \c_hash_str \q_stop } { \c_hash_str }
+ {
+ \str_if_eq:eeT
+ {#1}
+ { \c_hash_str \c_space_tl Conditional~Mappings }
+ { \ior_map_break: }
+ }
+ { \@@_data_auxi:w #1 \q_stop }
+ }
+ \ior_close:N \g_@@_data_ior
+\group_end:
+% \end{macrocode}
+%
+% \begin{macro}[EXP]{\__kernel_codepoint_case:nn}
+% \begin{macro}[EXP]{\@@_case:nnn}
+% \begin{macro}[EXP]
+% {\@@_uppercase:n, \@@_lowercase:n, \@@_titlecase:n, \@@_casefold:n}
+% \begin{macro}[EXP]{\@@_case:nn}
+% With the core data files loaded, there is now a need to provide access to
+% this information for other modules. That is done here such that case
+% folding can also be covered. At this level, all that needs to be returned
+% is the
+% \begin{macrocode}
+\cs_new:Npn \__kernel_codepoint_case:nn #1#2
+ {
+ \exp_args:Ne \@@_case:nnn
+ { \codepoint_str_generate:n {#2} } {#1} {#2}
+ }
+\cs_new:Npn \@@_case:nnn #1#2#3
+ {
+ \cs_if_exist:cTF { c_@@_ #2 _ #1 _tl }
+ {
+ \tl_use:c
+ { c_@@_ #2 _ #1 _tl }
+ }
+ { \use:c { @@_ #2 :n } {#3} }
+ }
+\cs_new:Npn \@@_uppercase:n { \@@_case:nn { uppercase } }
+\cs_new:Npn \@@_lowercase:n { \@@_case:nn { lowercase } }
+\cs_new:Npn \@@_titlecase:n { \@@_case:nn { uppercase } }
+\cs_new:Npn \@@_casefold:n { \@@_case:nn { lowercase } }
+\cs_new:Npn \@@_case:nn #1#2
+ {
+ { \int_eval:n { \__kernel_codepoint_data:nn {#1} {#2} + #2 } }
+ { }
+ { }
+ }
+% \end{macrocode}
+% \end{macro}
+% \end{macro}
+% \end{macro}
+% \end{macro}
+%
+% \begin{macro}[EXP]{\@@_nfd:n}
+% \begin{macro}[EXP]{\@@_nfd:nn}
+% A simple interface.
+% \begin{macrocode}
+\cs_new:Npn \@@_nfd:n #1
+ { \exp_args:Ne \@@_nfd:nn { \codepoint_str_generate:n {#1} } {#1} }
+\cs_new:Npn \@@_nfd:nn #1#2
+ {
+ \tl_if_exist:cTF { c_@@_nfd_ #1 _tl }
+ { \tl_use:c { c_@@_nfd_ #1 _tl } }
+ { {#2} { } }
+ }
+% \end{macrocode}
+% \end{macro}
+% \end{macro}
+%
+% \begin{macrocode}
+%<@@=text>
+% \end{macrocode}
+%
+% Read the Unicode grapheme data. This is quite easy to handle and we only need
+% codepoints, not characters, so there is no need to worry about the engine in use.
+% As reading as a string is most convenient, we have to do some work to remove
+% spaces: the hardest part of the entire process!
+% \begin{macrocode}
+\ior_new:N \g_@@_data_ior
+\group_begin:
+ \ior_open:Nn \g_@@_data_ior { GraphemeBreakProperty.txt }
+ \cs_set_nopar:Npn \l_@@_tmpa_str { }
+ \cs_set_nopar:Npn \l_@@_tmpb_str { }
+ \cs_set_protected:Npn \@@_data_auxi:w #1 ;~ #2 ~ #3 \q_stop
+ {
+ \str_if_eq:VnF \l_@@_tmpb_str {#2}
+ {
+ \str_if_empty:NF \l_@@_tmpb_str
+ {
+ \clist_const:ce { c_@@_grapheme_ \l_@@_tmpb_str _clist }
+ { \exp_after:wN \use_none:n \l_@@_tmpa_str }
+ \cs_set_nopar:Npn \l_@@_tmpa_str { }
+ }
+ \cs_set_nopar:Npn \l_@@_tmpb_str {#2}
+ }
+ \@@_data_auxii:w #1 .. #1 .. #1 \q_stop
+ }
+ \cs_set_protected:Npn \@@_data_auxii:w #1 .. #2 .. #3 \q_stop
+ {
+ \cs_set_nopar:Npe \l_@@_tmpa_str
+ {
+ \l_@@_tmpa_str ,
+ \tl_trim_spaces:n {#1} .. \tl_trim_spaces:n {#2}
+ }
+ }
+ \ior_str_map_inline:Nn \g_@@_data_ior
+ {
+ \str_if_eq:eeF { \tl_head:w #1 \c_hash_str \q_stop } { \c_hash_str }
+ {
+ \tl_if_blank:nF {#1}
+ { \@@_data_auxi:w #1 \q_stop }
+ }
+ }
+ \ior_close:N \g_@@_data_ior
+\group_end:
+% \end{macrocode}
+%
+% \begin{macrocode}
+%</package>
+% \end{macrocode}
+%
+% \end{implementation}
+%
+% \PrintIndex