Introduce nvim namespace: Move files.

Move files from src/ to src/nvim/.
- src/nvim/ becomes the new root dir for nvim executable sources.
- src/libnvim/ is planned to become root dir of the neovim library.

1 files changed, 13611 insertions, 0 deletions

diff --git a/src/nvim/spell.c b/src/nvim/spell.c
new file mode 100644
index 0000000000..5392fe4a70
--- /dev/null
+++ b/src/nvim/spell.c
@@ -0,0 +1,13611 @@
+// VIM - Vi IMproved    by Bram Moolenaar
+//
+// Do ":help uganda"  in Vim to read copying and usage conditions.
+// Do ":help credits" in Vim to see a list of people who contributed.
+// See README.txt for an overview of the Vim source code.
+
+// spell.c: code for spell checking
+//
+// The spell checking mechanism uses a tree (aka trie).  Each node in the tree
+// has a list of bytes that can appear (siblings).  For each byte there is a
+// pointer to the node with the byte that follows in the word (child).
+//
+// A NUL byte is used where the word may end.  The bytes are sorted, so that
+// binary searching can be used and the NUL bytes are at the start.  The
+// number of possible bytes is stored before the list of bytes.
+//
+// The tree uses two arrays: "byts" stores the characters, "idxs" stores
+// either the next index or flags.  The tree starts at index 0.  For example,
+// to lookup "vi" this sequence is followed:
+//      i = 0
+//      len = byts[i]
+//      n = where "v" appears in byts[i + 1] to byts[i + len]
+//      i = idxs[n]
+//      len = byts[i]
+//      n = where "i" appears in byts[i + 1] to byts[i + len]
+//      i = idxs[n]
+//      len = byts[i]
+//      find that byts[i + 1] is 0, idxs[i + 1] has flags for "vi".
+//
+// There are two word trees: one with case-folded words and one with words in
+// original case.  The second one is only used for keep-case words and is
+// usually small.
+//
+// There is one additional tree for when not all prefixes are applied when
+// generating the .spl file.  This tree stores all the possible prefixes, as
+// if they were words.  At each word (prefix) end the prefix nr is stored, the
+// following word must support this prefix nr.  And the condition nr is
+// stored, used to lookup the condition that the word must match with.
+//
+// Thanks to Olaf Seibert for providing an example implementation of this tree
+// and the compression mechanism.
+// LZ trie ideas:
+//      http://www.irb.hr/hr/home/ristov/papers/RistovLZtrieRevision1.pdf
+// More papers: http://www-igm.univ-mlv.fr/~laporte/publi_en.html
+//
+// Matching involves checking the caps type: Onecap ALLCAP KeepCap.
+//
+// Why doesn't Vim use aspell/ispell/myspell/etc.?
+// See ":help develop-spell".
+
+// Use SPELL_PRINTTREE for debugging: dump the word tree after adding a word.
+// Only use it for small word lists!
+
+// Use DEBUG_TRIEWALK to print the changes made in suggest_trie_walk() for a
+// specific word.
+
+// Use this to adjust the score after finding suggestions, based on the
+// suggested word sounding like the bad word.  This is much faster than doing
+// it for every possible suggestion.
+// Disadvantage: When "the" is typed as "hte" it sounds quite different ("@"
+// vs "ht") and goes down in the list.
+// Used when 'spellsuggest' is set to "best".
+#define RESCORE(word_score, sound_score) ((3 * word_score + sound_score) / 4)
+
+// Do the opposite: based on a maximum end score and a known sound score,
+// compute the maximum word score that can be used.
+#define MAXSCORE(word_score, sound_score) ((4 * word_score - sound_score) / 3)
+
+// Vim spell file format: <HEADER>
+//                        <SECTIONS>
+//                        <LWORDTREE>
+//                        <KWORDTREE>
+//                        <PREFIXTREE>
+//
+// <HEADER>: <fileID> <versionnr>
+//
+// <fileID>     8 bytes    "VIMspell"
+// <versionnr>  1 byte      VIMSPELLVERSION
+//
+//
+// Sections make it possible to add information to the .spl file without
+// making it incompatible with previous versions.  There are two kinds of
+// sections:
+// 1. Not essential for correct spell checking.  E.g. for making suggestions.
+//    These are skipped when not supported.
+// 2. Optional information, but essential for spell checking when present.
+//    E.g. conditions for affixes.  When this section is present but not
+//    supported an error message is given.
+//
+// <SECTIONS>: <section> ... <sectionend>
+//
+// <section>: <sectionID> <sectionflags> <sectionlen> (section contents)
+//
+// <sectionID>    1 byte    number from 0 to 254 identifying the section
+//
+// <sectionflags> 1 byte    SNF_REQUIRED: this section is required for correct
+//                                          spell checking
+//
+// <sectionlen>   4 bytes   length of section contents, MSB first
+//
+// <sectionend>   1 byte    SN_END
+//
+//
+// sectionID == SN_INFO: <infotext>
+// <infotext>    N bytes    free format text with spell file info (version,
+//                          website, etc)
+//
+// sectionID == SN_REGION: <regionname> ...
+// <regionname>  2 bytes    Up to 8 region names: ca, au, etc.  Lower case.
+//                          First <regionname> is region 1.
+//
+// sectionID == SN_CHARFLAGS: <charflagslen> <charflags>
+//                              <folcharslen> <folchars>
+// <charflagslen> 1 byte    Number of bytes in <charflags> (should be 128).
+// <charflags>  N bytes     List of flags (first one is for character 128):
+//                          0x01  word character        CF_WORD
+//                          0x02  upper-case character  CF_UPPER
+// <folcharslen>  2 bytes   Number of bytes in <folchars>.
+// <folchars>     N bytes   Folded characters, first one is for character 128.
+//
+// sectionID == SN_MIDWORD: <midword>
+// <midword>     N bytes    Characters that are word characters only when used
+//                          in the middle of a word.
+//
+// sectionID == SN_PREFCOND: <prefcondcnt> <prefcond> ...
+// <prefcondcnt> 2 bytes    Number of <prefcond> items following.
+// <prefcond> : <condlen> <condstr>
+// <condlen>    1 byte      Length of <condstr>.
+// <condstr>    N bytes     Condition for the prefix.
+//
+// sectionID == SN_REP: <repcount> <rep> ...
+// <repcount>    2 bytes    number of <rep> items, MSB first.
+// <rep> : <repfromlen> <repfrom> <reptolen> <repto>
+// <repfromlen>  1 byte     length of <repfrom>
+// <repfrom>     N bytes    "from" part of replacement
+// <reptolen>    1 byte     length of <repto>
+// <repto>       N bytes    "to" part of replacement
+//
+// sectionID == SN_REPSAL: <repcount> <rep> ...
+//   just like SN_REP but for soundfolded words
+//
+// sectionID == SN_SAL: <salflags> <salcount> <sal> ...
+// <salflags>    1 byte     flags for soundsalike conversion:
+//                          SAL_F0LLOWUP
+//                          SAL_COLLAPSE
+//                          SAL_REM_ACCENTS
+// <salcount>    2 bytes    number of <sal> items following
+// <sal> : <salfromlen> <salfrom> <saltolen> <salto>
+// <salfromlen>  1 byte     length of <salfrom>
+// <salfrom>     N bytes    "from" part of soundsalike
+// <saltolen>    1 byte     length of <salto>
+// <salto>       N bytes    "to" part of soundsalike
+//
+// sectionID == SN_SOFO: <sofofromlen> <sofofrom> <sofotolen> <sofoto>
+// <sofofromlen> 2 bytes    length of <sofofrom>
+// <sofofrom>    N bytes    "from" part of soundfold
+// <sofotolen>   2 bytes    length of <sofoto>
+// <sofoto>      N bytes    "to" part of soundfold
+//
+// sectionID == SN_SUGFILE: <timestamp>
+// <timestamp>   8 bytes    time in seconds that must match with .sug file
+//
+// sectionID == SN_NOSPLITSUGS: nothing
+//
+// sectionID == SN_WORDS: <word> ...
+// <word>        N bytes    NUL terminated common word
+//
+// sectionID == SN_MAP: <mapstr>
+// <mapstr>      N bytes    String with sequences of similar characters,
+//                          separated by slashes.
+//
+// sectionID == SN_COMPOUND: <compmax> <compminlen> <compsylmax> <compoptions>
+//                              <comppatcount> <comppattern> ... <compflags>
+// <compmax>     1 byte     Maximum nr of words in compound word.
+// <compminlen>  1 byte     Minimal word length for compounding.
+// <compsylmax>  1 byte     Maximum nr of syllables in compound word.
+// <compoptions> 2 bytes    COMP_ flags.
+// <comppatcount> 2 bytes   number of <comppattern> following
+// <compflags>   N bytes    Flags from COMPOUNDRULE items, separated by
+//                          slashes.
+//
+// <comppattern>: <comppatlen> <comppattext>
+// <comppatlen>  1 byte     length of <comppattext>
+// <comppattext> N bytes    end or begin chars from CHECKCOMPOUNDPATTERN
+//
+// sectionID == SN_NOBREAK: (empty, its presence is what matters)
+//
+// sectionID == SN_SYLLABLE: <syllable>
+// <syllable>    N bytes    String from SYLLABLE item.
+//
+// <LWORDTREE>: <wordtree>
+//
+// <KWORDTREE>: <wordtree>
+//
+// <PREFIXTREE>: <wordtree>
+//
+//
+// <wordtree>: <nodecount> <nodedata> ...
+//
+// <nodecount>  4 bytes     Number of nodes following.  MSB first.
+//
+// <nodedata>: <siblingcount> <sibling> ...
+//
+// <siblingcount> 1 byte    Number of siblings in this node.  The siblings
+//                          follow in sorted order.
+//
+// <sibling>: <byte> [ <nodeidx> <xbyte>
+//                    | <flags> [<flags2>] [<region>] [<affixID>]
+//                    | [<pflags>] <affixID> <prefcondnr> ]
+//
+// <byte>       1 byte      Byte value of the sibling.  Special cases:
+//                          BY_NOFLAGS: End of word without flags and for all
+//                                      regions.
+//                                      For PREFIXTREE <affixID> and
+//                                      <prefcondnr> follow.
+//                          BY_FLAGS:   End of word, <flags> follow.
+//                                      For PREFIXTREE <pflags>, <affixID>
+//                                      and <prefcondnr> follow.
+//                          BY_FLAGS2:  End of word, <flags> and <flags2>
+//                                      follow.  Not used in PREFIXTREE.
+//                          BY_INDEX:   Child of sibling is shared, <nodeidx>
+//                                      and <xbyte> follow.
+//
+// <nodeidx>    3 bytes     Index of child for this sibling, MSB first.
+//
+// <xbyte>      1 byte      Byte value of the sibling.
+//
+// <flags>      1 byte      Bitmask of:
+//                          WF_ALLCAP   word must have only capitals
+//                          WF_ONECAP   first char of word must be capital
+//                          WF_KEEPCAP  keep-case word
+//                          WF_FIXCAP   keep-case word, all caps not allowed
+//                          WF_RARE     rare word
+//                          WF_BANNED   bad word
+//                          WF_REGION   <region> follows
+//                          WF_AFX      <affixID> follows
+//
+// <flags2>     1 byte      Bitmask of:
+//                          WF_HAS_AFF >> 8   word includes affix
+//                          WF_NEEDCOMP >> 8  word only valid in compound
+//                          WF_NOSUGGEST >> 8  word not used for suggestions
+//                          WF_COMPROOT >> 8  word already a compound
+//                          WF_NOCOMPBEF >> 8 no compounding before this word
+//                          WF_NOCOMPAFT >> 8 no compounding after this word
+//
+// <pflags>     1 byte      Bitmask of:
+//                          WFP_RARE    rare prefix
+//                          WFP_NC      non-combining prefix
+//                          WFP_UP      letter after prefix made upper case
+//
+// <region>     1 byte      Bitmask for regions in which word is valid.  When
+//                          omitted it's valid in all regions.
+//                          Lowest bit is for region 1.
+//
+// <affixID>    1 byte      ID of affix that can be used with this word.  In
+//                          PREFIXTREE used for the required prefix ID.
+//
+// <prefcondnr> 2 bytes     Prefix condition number, index in <prefcond> list
+//                          from HEADER.
+//
+// All text characters are in 'encoding', but stored as single bytes.
+
+// Vim .sug file format:  <SUGHEADER>
+//                        <SUGWORDTREE>
+//                        <SUGTABLE>
+//
+// <SUGHEADER>: <fileID> <versionnr> <timestamp>
+//
+// <fileID>     6 bytes     "VIMsug"
+// <versionnr>  1 byte      VIMSUGVERSION
+// <timestamp>  8 bytes     timestamp that must match with .spl file
+//
+//
+// <SUGWORDTREE>: <wordtree>  (see above, no flags or region used)
+//
+//
+// <SUGTABLE>: <sugwcount> <sugline> ...
+//
+// <sugwcount>  4 bytes     number of <sugline> following
+//
+// <sugline>: <sugnr> ... NUL
+//
+// <sugnr>:     X bytes     word number that results in this soundfolded word,
+//                          stored as an offset to the previous number in as
+//                          few bytes as possible, see offset2bytes())
+
+#include <string.h>
+#include <stdlib.h>
+
+#include "vim.h"
+#include "spell.h"
+#include "buffer.h"
+#include "charset.h"
+#include "edit.h"
+#include "eval.h"
+#include "ex_cmds.h"
+#include "ex_cmds2.h"
+#include "ex_docmd.h"
+#include "fileio.h"
+#include "getchar.h"
+#include "hashtab.h"
+#include "mbyte.h"
+#include "memline.h"
+#include "memory.h"
+#include "message.h"
+#include "misc1.h"
+#include "misc2.h"
+#include "garray.h"
+#include "normal.h"
+#include "option.h"
+#include "os_unix.h"
+#include "path.h"
+#include "regexp.h"
+#include "screen.h"
+#include "search.h"
+#include "syntax.h"
+#include "term.h"
+#include "ui.h"
+#include "undo.h"
+#include "os/os.h"
+
+#ifndef UNIX            // it's in os_unix_defs.h for Unix
+# include <time.h>      // for time_t
+#endif
+
+#define MAXWLEN 250             // Assume max. word len is this many bytes.
+                                // Some places assume a word length fits in a
+                                // byte, thus it can't be above 255.
+
+// Type used for indexes in the word tree need to be at least 4 bytes.  If int
+// is 8 bytes we could use something smaller, but what?
+typedef int idx_T;
+
+# define SPL_FNAME_TMPL  "%s.%s.spl"
+# define SPL_FNAME_ADD   ".add."
+# define SPL_FNAME_ASCII ".ascii."
+
+// Flags used for a word.  Only the lowest byte can be used, the region byte
+// comes above it.
+#define WF_REGION   0x01        // region byte follows
+#define WF_ONECAP   0x02        // word with one capital (or all capitals)
+#define WF_ALLCAP   0x04        // word must be all capitals
+#define WF_RARE     0x08        // rare word
+#define WF_BANNED   0x10        // bad word
+#define WF_AFX      0x20        // affix ID follows
+#define WF_FIXCAP   0x40        // keep-case word, allcap not allowed
+#define WF_KEEPCAP  0x80        // keep-case word
+
+// for <flags2>, shifted up one byte to be used in wn_flags
+#define WF_HAS_AFF  0x0100      // word includes affix
+#define WF_NEEDCOMP 0x0200      // word only valid in compound
+#define WF_NOSUGGEST 0x0400     // word not to be suggested
+#define WF_COMPROOT 0x0800      // already compounded word, COMPOUNDROOT
+#define WF_NOCOMPBEF 0x1000     // no compounding before this word
+#define WF_NOCOMPAFT 0x2000     // no compounding after this word
+
+// only used for su_badflags
+#define WF_MIXCAP   0x20        // mix of upper and lower case: macaRONI
+
+#define WF_CAPMASK (WF_ONECAP | WF_ALLCAP | WF_KEEPCAP | WF_FIXCAP)
+
+// flags for <pflags>
+#define WFP_RARE            0x01        // rare prefix
+#define WFP_NC              0x02        // prefix is not combining
+#define WFP_UP              0x04        // to-upper prefix
+#define WFP_COMPPERMIT      0x08        // prefix with COMPOUNDPERMITFLAG
+#define WFP_COMPFORBID      0x10        // prefix with COMPOUNDFORBIDFLAG
+
+// Flags for postponed prefixes in "sl_pidxs".  Must be above affixID (one
+// byte) and prefcondnr (two bytes).
+#define WF_RAREPFX  (WFP_RARE << 24)    // rare postponed prefix
+#define WF_PFX_NC   (WFP_NC << 24)      // non-combining postponed prefix
+#define WF_PFX_UP   (WFP_UP << 24)      // to-upper postponed prefix
+#define WF_PFX_COMPPERMIT (WFP_COMPPERMIT << 24) // postponed prefix with
+                                                 // COMPOUNDPERMITFLAG
+#define WF_PFX_COMPFORBID (WFP_COMPFORBID << 24) // postponed prefix with
+                                                 // COMPOUNDFORBIDFLAG
+
+
+// flags for <compoptions>
+#define COMP_CHECKDUP           1       // CHECKCOMPOUNDDUP
+#define COMP_CHECKREP           2       // CHECKCOMPOUNDREP
+#define COMP_CHECKCASE          4       // CHECKCOMPOUNDCASE
+#define COMP_CHECKTRIPLE        8       // CHECKCOMPOUNDTRIPLE
+
+// Special byte values for <byte>.  Some are only used in the tree for
+// postponed prefixes, some only in the other trees.  This is a bit messy...
+#define BY_NOFLAGS      0       // end of word without flags or region; for
+                                // postponed prefix: no <pflags>
+#define BY_INDEX        1       // child is shared, index follows
+#define BY_FLAGS        2       // end of word, <flags> byte follows; for
+                                // postponed prefix: <pflags> follows
+#define BY_FLAGS2       3       // end of word, <flags> and <flags2> bytes
+                                // follow; never used in prefix tree
+#define BY_SPECIAL  BY_FLAGS2   // highest special byte value
+
+// Info from "REP", "REPSAL" and "SAL" entries in ".aff" file used in si_rep,
+// si_repsal, sl_rep, and si_sal.  Not for sl_sal!
+// One replacement: from "ft_from" to "ft_to".
+typedef struct fromto_S {
+  char_u      *ft_from;
+  char_u      *ft_to;
+} fromto_T;
+
+// Info from "SAL" entries in ".aff" file used in sl_sal.
+// The info is split for quick processing by spell_soundfold().
+// Note that "sm_oneof" and "sm_rules" point into sm_lead.
+typedef struct salitem_S {
+  char_u      *sm_lead;         // leading letters
+  int sm_leadlen;               // length of "sm_lead"
+  char_u      *sm_oneof;        // letters from () or NULL
+  char_u      *sm_rules;        // rules like ^, $, priority
+  char_u      *sm_to;           // replacement.
+  int         *sm_lead_w;       // wide character copy of "sm_lead"
+  int         *sm_oneof_w;      // wide character copy of "sm_oneof"
+  int         *sm_to_w;         // wide character copy of "sm_to"
+} salitem_T;
+
+typedef int salfirst_T;
+
+// Values for SP_*ERROR are negative, positive values are used by
+// read_cnt_string().
+#define SP_TRUNCERROR   -1      // spell file truncated error
+#define SP_FORMERROR    -2      // format error in spell file
+#define SP_OTHERERROR   -3      // other error while reading spell file
+
+// Structure used to store words and other info for one language, loaded from
+// a .spl file.
+// The main access is through the tree in "sl_fbyts/sl_fidxs", storing the
+// case-folded words.  "sl_kbyts/sl_kidxs" is for keep-case words.
+//
+// The "byts" array stores the possible bytes in each tree node, preceded by
+// the number of possible bytes, sorted on byte value:
+//      <len> <byte1> <byte2> ...
+// The "idxs" array stores the index of the child node corresponding to the
+// byte in "byts".
+// Exception: when the byte is zero, the word may end here and "idxs" holds
+// the flags, region mask and affixID for the word.  There may be several
+// zeros in sequence for alternative flag/region/affixID combinations.
+typedef struct slang_S slang_T;
+struct slang_S {
+  slang_T     *sl_next;         // next language
+  char_u      *sl_name;         // language name "en", "en.rare", "nl", etc.
+  char_u      *sl_fname;        // name of .spl file
+  int sl_add;                   // TRUE if it's a .add file.
+
+  char_u      *sl_fbyts;        // case-folded word bytes
+  idx_T       *sl_fidxs;        // case-folded word indexes
+  char_u      *sl_kbyts;        // keep-case word bytes
+  idx_T       *sl_kidxs;        // keep-case word indexes
+  char_u      *sl_pbyts;        // prefix tree word bytes
+  idx_T       *sl_pidxs;        // prefix tree word indexes
+
+  char_u      *sl_info;         // infotext string or NULL
+
+  char_u sl_regions[17];        // table with up to 8 region names plus NUL
+
+  char_u      *sl_midword;      // MIDWORD string or NULL
+
+  hashtab_T sl_wordcount;       // hashtable with word count, wordcount_T
+
+  int sl_compmax;               // COMPOUNDWORDMAX (default: MAXWLEN)
+  int sl_compminlen;            // COMPOUNDMIN (default: 0)
+  int sl_compsylmax;            // COMPOUNDSYLMAX (default: MAXWLEN)
+  int sl_compoptions;           // COMP_* flags
+  garray_T sl_comppat;          // CHECKCOMPOUNDPATTERN items
+  regprog_T   *sl_compprog;     // COMPOUNDRULE turned into a regexp progrm
+                                // (NULL when no compounding)
+  char_u      *sl_comprules;    // all COMPOUNDRULE concatenated (or NULL)
+  char_u      *sl_compstartflags;   // flags for first compound word
+  char_u      *sl_compallflags;   // all flags for compound words
+  char_u sl_nobreak;            // When TRUE: no spaces between words
+  char_u      *sl_syllable;     // SYLLABLE repeatable chars or NULL
+  garray_T sl_syl_items;        // syllable items
+
+  int sl_prefixcnt;             // number of items in "sl_prefprog"
+  regprog_T   **sl_prefprog;    // table with regprogs for prefixes
+
+  garray_T sl_rep;              // list of fromto_T entries from REP lines
+  short sl_rep_first[256];          // indexes where byte first appears, -1 if
+                                    // there is none
+  garray_T sl_sal;              // list of salitem_T entries from SAL lines
+  salfirst_T sl_sal_first[256];     // indexes where byte first appears, -1 if
+                                    // there is none
+  int sl_followup;              // SAL followup
+  int sl_collapse;              // SAL collapse_result
+  int sl_rem_accents;           // SAL remove_accents
+  int sl_sofo;                  // SOFOFROM and SOFOTO instead of SAL items:
+                                // "sl_sal_first" maps chars, when has_mbyte
+                                // "sl_sal" is a list of wide char lists.
+  garray_T sl_repsal;           // list of fromto_T entries from REPSAL lines
+  short sl_repsal_first[256];   // sl_rep_first for REPSAL lines
+  int sl_nosplitsugs;           // don't suggest splitting a word
+
+  // Info from the .sug file.  Loaded on demand.
+  time_t sl_sugtime;            // timestamp for .sug file
+  char_u      *sl_sbyts;        // soundfolded word bytes
+  idx_T       *sl_sidxs;        // soundfolded word indexes
+  buf_T       *sl_sugbuf;       // buffer with word number table
+  int sl_sugloaded;             // TRUE when .sug file was loaded or failed to
+                                // load
+
+  int sl_has_map;               // TRUE if there is a MAP line
+  hashtab_T sl_map_hash;        // MAP for multi-byte chars
+  int sl_map_array[256];        // MAP for first 256 chars
+  hashtab_T sl_sounddone;       // table with soundfolded words that have
+                                // handled, see add_sound_suggest()
+};
+
+// First language that is loaded, start of the linked list of loaded
+// languages.
+static slang_T *first_lang = NULL;
+
+// Flags used in .spl file for soundsalike flags.
+#define SAL_F0LLOWUP            1
+#define SAL_COLLAPSE            2
+#define SAL_REM_ACCENTS         4
+
+// Structure used in "b_langp", filled from 'spelllang'.
+typedef struct langp_S {
+  slang_T     *lp_slang;        // info for this language
+  slang_T     *lp_sallang;      // language used for sound folding or NULL
+  slang_T     *lp_replang;      // language used for REP items or NULL
+  int lp_region;                // bitmask for region or REGION_ALL
+} langp_T;
+
+#define LANGP_ENTRY(ga, i)      (((langp_T *)(ga).ga_data) + (i))
+
+#define REGION_ALL 0xff         // word valid in all regions
+
+#define VIMSPELLMAGIC "VIMspell"  // string at start of Vim spell file
+#define VIMSPELLMAGICL 8
+#define VIMSPELLVERSION 50
+
+#define VIMSUGMAGIC "VIMsug"    // string at start of Vim .sug file
+#define VIMSUGMAGICL 6
+#define VIMSUGVERSION 1
+
+// Section IDs.  Only renumber them when VIMSPELLVERSION changes!
+#define SN_REGION       0       // <regionname> section
+#define SN_CHARFLAGS    1       // charflags section
+#define SN_MIDWORD      2       // <midword> section
+#define SN_PREFCOND     3       // <prefcond> section
+#define SN_REP          4       // REP items section
+#define SN_SAL          5       // SAL items section
+#define SN_SOFO         6       // soundfolding section
+#define SN_MAP          7       // MAP items section
+#define SN_COMPOUND     8       // compound words section
+#define SN_SYLLABLE     9       // syllable section
+#define SN_NOBREAK      10      // NOBREAK section
+#define SN_SUGFILE      11      // timestamp for .sug file
+#define SN_REPSAL       12      // REPSAL items section
+#define SN_WORDS        13      // common words
+#define SN_NOSPLITSUGS  14      // don't split word for suggestions
+#define SN_INFO         15      // info section
+#define SN_END          255     // end of sections
+
+#define SNF_REQUIRED    1       // <sectionflags>: required section
+
+// Result values.  Lower number is accepted over higher one.
+#define SP_BANNED       -1
+#define SP_OK           0
+#define SP_RARE         1
+#define SP_LOCAL        2
+#define SP_BAD          3
+
+// file used for "zG" and "zW"
+static char_u   *int_wordlist = NULL;
+
+typedef struct wordcount_S {
+  uint16_t wc_count;                // nr of times word was seen
+  char_u wc_word[1];                // word, actually longer
+} wordcount_T;
+
+static wordcount_T dumwc;
+#define WC_KEY_OFF  (unsigned)(dumwc.wc_word - (char_u *)&dumwc)
+#define HI2WC(hi)     ((wordcount_T *)((hi)->hi_key - WC_KEY_OFF))
+#define MAXWORDCOUNT 0xffff
+
+// Information used when looking for suggestions.
+typedef struct suginfo_S {
+  garray_T su_ga;                   // suggestions, contains "suggest_T"
+  int su_maxcount;                  // max. number of suggestions displayed
+  int su_maxscore;                  // maximum score for adding to su_ga
+  int su_sfmaxscore;                // idem, for when doing soundfold words
+  garray_T su_sga;                  // like su_ga, sound-folded scoring
+  char_u      *su_badptr;           // start of bad word in line
+  int su_badlen;                    // length of detected bad word in line
+  int su_badflags;                  // caps flags for bad word
+  char_u su_badword[MAXWLEN];       // bad word truncated at su_badlen
+  char_u su_fbadword[MAXWLEN];      // su_badword case-folded
+  char_u su_sal_badword[MAXWLEN];   // su_badword soundfolded
+  hashtab_T su_banned;              // table with banned words
+  slang_T     *su_sallang;          // default language for sound folding
+} suginfo_T;
+
+// One word suggestion.  Used in "si_ga".
+typedef struct suggest_S {
+  char_u      *st_word;         // suggested word, allocated string
+  int st_wordlen;               // STRLEN(st_word)
+  int st_orglen;                // length of replaced text
+  int st_score;                 // lower is better
+  int st_altscore;              // used when st_score compares equal
+  int st_salscore;              // st_score is for soundalike
+  int st_had_bonus;             // bonus already included in score
+  slang_T     *st_slang;        // language used for sound folding
+} suggest_T;
+
+#define SUG(ga, i) (((suggest_T *)(ga).ga_data)[i])
+
+// TRUE if a word appears in the list of banned words.
+#define WAS_BANNED(su, word) (!HASHITEM_EMPTY(hash_find(&su->su_banned, word)))
+
+// Number of suggestions kept when cleaning up.  We need to keep more than
+// what is displayed, because when rescore_suggestions() is called the score
+// may change and wrong suggestions may be removed later.
+#define SUG_CLEAN_COUNT(su)    ((su)->su_maxcount < \
+                                130 ? 150 : (su)->su_maxcount + 20)
+
+// Threshold for sorting and cleaning up suggestions.  Don't want to keep lots
+// of suggestions that are not going to be displayed.
+#define SUG_MAX_COUNT(su)       (SUG_CLEAN_COUNT(su) + 50)
+
+// score for various changes
+#define SCORE_SPLIT     149     // split bad word
+#define SCORE_SPLIT_NO  249     // split bad word with NOSPLITSUGS
+#define SCORE_ICASE     52      // slightly different case
+#define SCORE_REGION    200     // word is for different region
+#define SCORE_RARE      180     // rare word
+#define SCORE_SWAP      75      // swap two characters
+#define SCORE_SWAP3     110     // swap two characters in three
+#define SCORE_REP       65      // REP replacement
+#define SCORE_SUBST     93      // substitute a character
+#define SCORE_SIMILAR   33      // substitute a similar character
+#define SCORE_SUBCOMP   33      // substitute a composing character
+#define SCORE_DEL       94      // delete a character
+#define SCORE_DELDUP    66      // delete a duplicated character
+#define SCORE_DELCOMP   28      // delete a composing character
+#define SCORE_INS       96      // insert a character
+#define SCORE_INSDUP    67      // insert a duplicate character
+#define SCORE_INSCOMP   30      // insert a composing character
+#define SCORE_NONWORD   103     // change non-word to word char
+
+#define SCORE_FILE      30      // suggestion from a file
+#define SCORE_MAXINIT   350     // Initial maximum score: higher == slower.
+                                // 350 allows for about three changes.
+
+#define SCORE_COMMON1   30      // subtracted for words seen before
+#define SCORE_COMMON2   40      // subtracted for words often seen
+#define SCORE_COMMON3   50      // subtracted for words very often seen
+#define SCORE_THRES2    10      // word count threshold for COMMON2
+#define SCORE_THRES3    100     // word count threshold for COMMON3
+
+// When trying changed soundfold words it becomes slow when trying more than
+// two changes.  With less then two changes it's slightly faster but we miss a
+// few good suggestions.  In rare cases we need to try three of four changes.
+#define SCORE_SFMAX1    200     // maximum score for first try
+#define SCORE_SFMAX2    300     // maximum score for second try
+#define SCORE_SFMAX3    400     // maximum score for third try
+
+#define SCORE_BIG       SCORE_INS * 3   // big difference
+#define SCORE_MAXMAX    999999          // accept any score
+#define SCORE_LIMITMAX  350             // for spell_edit_score_limit()
+
+// for spell_edit_score_limit() we need to know the minimum value of
+// SCORE_ICASE, SCORE_SWAP, SCORE_DEL, SCORE_SIMILAR and SCORE_INS
+#define SCORE_EDIT_MIN  SCORE_SIMILAR
+
+// Structure to store info for word matching.
+typedef struct matchinf_S {
+  langp_T     *mi_lp;                   // info for language and region
+
+  // pointers to original text to be checked
+  char_u      *mi_word;                 // start of word being checked
+  char_u      *mi_end;                  // end of matching word so far
+  char_u      *mi_fend;                 // next char to be added to mi_fword
+  char_u      *mi_cend;                 // char after what was used for
+                                        // mi_capflags
+
+  // case-folded text
+  char_u mi_fword[MAXWLEN + 1];         // mi_word case-folded
+  int mi_fwordlen;                      // nr of valid bytes in mi_fword
+
+  // for when checking word after a prefix
+  int mi_prefarridx;                    // index in sl_pidxs with list of
+                                        // affixID/condition
+  int mi_prefcnt;                       // number of entries at mi_prefarridx
+  int mi_prefixlen;                     // byte length of prefix
+  int mi_cprefixlen;                    // byte length of prefix in original
+                                        // case
+
+  // for when checking a compound word
+  int mi_compoff;                       // start of following word offset
+  char_u mi_compflags[MAXWLEN];         // flags for compound words used
+  int mi_complen;                       // nr of compound words used
+  int mi_compextra;                     // nr of COMPOUNDROOT words
+
+  // others
+  int mi_result;                        // result so far: SP_BAD, SP_OK, etc.
+  int mi_capflags;                      // WF_ONECAP WF_ALLCAP WF_KEEPCAP
+  win_T       *mi_win;                  // buffer being checked
+
+  // for NOBREAK
+  int mi_result2;                       // "mi_resul" without following word
+  char_u      *mi_end2;                 // "mi_end" without following word
+} matchinf_T;
+
+// The tables used for recognizing word characters according to spelling.
+// These are only used for the first 256 characters of 'encoding'.
+typedef struct spelltab_S {
+  char_u st_isw[256];           // flags: is word char
+  char_u st_isu[256];           // flags: is uppercase char
+  char_u st_fold[256];        // chars: folded case
+  char_u st_upper[256];       // chars: upper case
+} spelltab_T;
+
+static spelltab_T spelltab;
+static int did_set_spelltab;
+
+#define CF_WORD         0x01
+#define CF_UPPER        0x02
+
+static void clear_spell_chartab(spelltab_T *sp);
+static int set_spell_finish(spelltab_T  *new_st);
+static int spell_iswordp(char_u *p, win_T *wp);
+static int spell_iswordp_nmw(char_u *p, win_T *wp);
+static int spell_mb_isword_class(int cl, win_T *wp);
+static int spell_iswordp_w(int *p, win_T *wp);
+static int write_spell_prefcond(FILE *fd, garray_T *gap);
+
+// For finding suggestions: At each node in the tree these states are tried:
+typedef enum {
+  STATE_START = 0,      // At start of node check for NUL bytes (goodword
+                        // ends); if badword ends there is a match, otherwise
+                        // try splitting word.
+  STATE_NOPREFIX,       // try without prefix
+  STATE_SPLITUNDO,      // Undo splitting.
+  STATE_ENDNUL,         // Past NUL bytes at start of the node.
+  STATE_PLAIN,          // Use each byte of the node.
+  STATE_DEL,            // Delete a byte from the bad word.
+  STATE_INS_PREP,       // Prepare for inserting bytes.
+  STATE_INS,            // Insert a byte in the bad word.
+  STATE_SWAP,           // Swap two bytes.
+  STATE_UNSWAP,         // Undo swap two characters.
+  STATE_SWAP3,          // Swap two characters over three.
+  STATE_UNSWAP3,        // Undo Swap two characters over three.
+  STATE_UNROT3L,        // Undo rotate three characters left
+  STATE_UNROT3R,        // Undo rotate three characters right
+  STATE_REP_INI,        // Prepare for using REP items.
+  STATE_REP,            // Use matching REP items from the .aff file.
+  STATE_REP_UNDO,       // Undo a REP item replacement.
+  STATE_FINAL           // End of this node.
+} state_T;
+
+// Struct to keep the state at each level in suggest_try_change().
+typedef struct trystate_S {
+  state_T ts_state;             // state at this level, STATE_
+  int ts_score;                 // score
+  idx_T ts_arridx;              // index in tree array, start of node
+  short ts_curi;                // index in list of child nodes
+  char_u ts_fidx;               // index in fword[], case-folded bad word
+  char_u ts_fidxtry;            // ts_fidx at which bytes may be changed
+  char_u ts_twordlen;           // valid length of tword[]
+  char_u ts_prefixdepth;        // stack depth for end of prefix or
+                                // PFD_PREFIXTREE or PFD_NOPREFIX
+  char_u ts_flags;              // TSF_ flags
+  char_u ts_tcharlen;           // number of bytes in tword character
+  char_u ts_tcharidx;           // current byte index in tword character
+  char_u ts_isdiff;             // DIFF_ values
+  char_u ts_fcharstart;         // index in fword where badword char started
+  char_u ts_prewordlen;         // length of word in "preword[]"
+  char_u ts_splitoff;           // index in "tword" after last split
+  char_u ts_splitfidx;          // "ts_fidx" at word split
+  char_u ts_complen;            // nr of compound words used
+  char_u ts_compsplit;          // index for "compflags" where word was spit
+  char_u ts_save_badflags;      // su_badflags saved here
+  char_u ts_delidx;             // index in fword for char that was deleted,
+                                // valid when "ts_flags" has TSF_DIDDEL
+} trystate_T;
+
+// values for ts_isdiff
+#define DIFF_NONE       0       // no different byte (yet)
+#define DIFF_YES        1       // different byte found
+#define DIFF_INSERT     2       // inserting character
+
+// values for ts_flags
+#define TSF_PREFIXOK    1       // already checked that prefix is OK
+#define TSF_DIDSPLIT    2       // tried split at this point
+#define TSF_DIDDEL      4       // did a delete, "ts_delidx" has index
+
+// special values ts_prefixdepth
+#define PFD_NOPREFIX    0xff    // not using prefixes
+#define PFD_PREFIXTREE  0xfe    // walking through the prefix tree
+#define PFD_NOTSPECIAL  0xfd    // highest value that's not special
+
+// mode values for find_word
+#define FIND_FOLDWORD       0   // find word case-folded
+#define FIND_KEEPWORD       1   // find keep-case word
+#define FIND_PREFIX         2   // find word after prefix
+#define FIND_COMPOUND       3   // find case-folded compound word
+#define FIND_KEEPCOMPOUND   4   // find keep-case compound word
+
+static slang_T *slang_alloc(char_u *lang);
+static void slang_free(slang_T *lp);
+static void slang_clear(slang_T *lp);
+static void slang_clear_sug(slang_T *lp);
+static void find_word(matchinf_T *mip, int mode);
+static int match_checkcompoundpattern(char_u *ptr, int wlen,
+                                      garray_T *gap);
+static int can_compound(slang_T *slang, char_u *word, char_u *flags);
+static int can_be_compound(trystate_T *sp, slang_T *slang, char_u *compflags,
+                           int flag);
+static int match_compoundrule(slang_T *slang, char_u *compflags);
+static int valid_word_prefix(int totprefcnt, int arridx, int flags,
+                             char_u *word, slang_T *slang,
+                             int cond_req);
+static void find_prefix(matchinf_T *mip, int mode);
+static int fold_more(matchinf_T *mip);