And fix the prefetches after \0.

And make sure that the unfilled entries are NULL to catch any spurious
reference.

Also fix some missing phoneme table switch according to language
changes.

And fill the last phlist prepause and newword fields, otherwise they are
detected as undefined:

==483407== Conditional jump or move depends on uninitialised value(s)
==483407==    at 0x488E6AB: Generate (synthesize.c:1228)
==483407==    by 0x488FD94: SpeakNextClause (synthesize.c:1587)
==483407==    by 0x4887F56: Synthesize (speech.c:457)
==483407==    by 0x488884C: sync_espeak_Synth (speech.c:570)
==483407==    by 0x487B270: espeak_Synth (espeak_api.c:90)
==483407==    by 0x10ACA0: main (espeak-ng.c:691)
==483407==  Uninitialised value was created by a client request
==483407==    at 0x4884893: MakePhonemeList (phonemelist.c:155)
==483407==    by 0x4895712: TranslateClause (translate.c:2682)
==483407==    by 0x488FCCF: SpeakNextClause (synthesize.c:1569)
==483407==    by 0x4887F56: Synthesize (speech.c:457)
==483407==    by 0x488884C: sync_espeak_Synth (speech.c:570)
==483407==    by 0x487B270: espeak_Synth (espeak_api.c:90)
==483407==    by 0x10ACA0: main (espeak-ng.c:691)
==483407==
==483407== Conditional jump or move depends on uninitialised value(s)
==483407==    at 0x488E622: Generate (synthesize.c:1211)
==483407==    by 0x488FD94: SpeakNextClause (synthesize.c:1587)
==483407==    by 0x4887F56: Synthesize (speech.c:457)
==483407==    by 0x488884C: sync_espeak_Synth (speech.c:570)
==483407==    by 0x487B270: espeak_Synth (espeak_api.c:90)
==483407==    by 0x10ACA0: main (espeak-ng.c:691)
==483407==  Uninitialised value was created by a client request
==483407==    at 0x4884893: MakePhonemeList (phonemelist.c:155)
==483407==    by 0x4895712: TranslateClause (translate.c:2682)
==483407==    by 0x488FCCF: SpeakNextClause (synthesize.c:1569)
==483407==    by 0x4887F56: Synthesize (speech.c:457)
==483407==    by 0x488884C: sync_espeak_Synth (speech.c:570)
==483407==    by 0x487B270: espeak_Synth (espeak_api.c:90)
==483407==    by 0x10ACA0: main (espeak-ng.c:691)
==483407==

This is changing the ssml.test output, but with no audible difference,
so this is probably a real fix for it.

And fix looking after end of array.

So they are protected from overflows, and shared between processes

These are global arrays reused several times. When using them msan and
valgrind thus believe they are always initialized, which reduces their
capacity to detect uninitialized values. We can however explicitly tell them
when they are reused, and thus to be considered as uninitialized.

When pollint() returns 100.0, multiplying by 2.55 doesn't actually seem to
be getting 255 on i386. Multiplying by 255 and dividing by 100, however,
does (probably because float computation with small integer values are
guaranteed to have integer results).

Fixes #1151

This has varying behavior depending on toolchain and processor.

Fixes #1151

When building espeak-ng in a directory that has a long path, 80 bytes to
store it is quite short.

Fixes #1152

For the case when word is smaller than 4 characters, we should not look at
the 3rd or 4th character before checking the previous ones, otherwise
we'd at best read uninitialized data, at worse non-existing data.

fname_temp is not always getting filled.

Some rules test against character not being of a certain type. That may
match with the \0 end-of-text marker, and thus actually step over
it and let MatchRule continue with uninitialized data after it, leading
to potential random behavior.

This commits fixes it by making sure that we don't read past that \0.

This seems to be changing the pronunciation of "capitals" from k'apIt@Lz to
k'apIt,alz, I don't know why, I guess the rule for it was actually
bogus?

Special characters such as N, S1, etc. are not actually eating
characters. Their treatment should thus *not* update pre_ptr and post_ptr,
otherwise those would underflow/overflow, e.g. in the case

@) s (_NS1  [z]

this would overflow. This for instance noticeable with the memory sanitizer:

ESPEAK_DATA_PATH=$PWD ./src/espeak-ng -qX "capitals"
Translate 'capitals'
  1	c        [k]

  1	a        [a]

  1	p        [p]

  1	i        [I]

  1	t        [t]

  1	a        [a]

  1	l        [l]
 20	l (C     [l]

==2837201==WARNING: MemorySanitizer: use-of-uninitialized-value
    #0 0x7f7f4422744b in utf8_in2 /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/translate.c:281:2
    #1 0x7f7f442281bc in utf8_in /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/translate.c:332:9
    #2 0x7f7f440e0d31 in MatchRule /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/dictionary.c:1767:21
    #3 0x7f7f440d937f in TranslateRules /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/dictionary.c:2320:6
    #4 0x7f7f44230e5f in TranslateWord3 /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/translate.c:733:15
    #5 0x7f7f44229844 in TranslateWord /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/translate.c:1100:14
    #6 0x7f7f44256e50 in TranslateWord2 /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/translate.c:1361:11
    #7 0x7f7f4424d6cc in TranslateClause /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/translate.c:2623:17
    #8 0x7f7f44213359 in SpeakNextClause /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/synthesize.c:1569:2
    #9 0x7f7f441a9f56 in Synthesize /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/speech.c:457:2
    #10 0x7f7f441a9023 in sync_espeak_Synth /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/speech.c:570:29
    #11 0x7f7f441ad59f in espeak_ng_Synthesize /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/speech.c:678:10
    #12 0x7f7f4410b3f4 in espeak_Synth /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/espeak_api.c:90:32
    #13 0x4a8be3 in main /home/samy/brl/speech/espeak-ng-git/src/espeak-ng.c:691:3
    #14 0x7f7f43a2e7fc in __libc_start_main csu/../csu/libc-start.c:332:16
    #15 0x421449 in _start (/home/samy/ens/projet/1/speech/espeak-ng-git/src/.libs/espeak-ng+0x421449)

  Uninitialized value was created by an allocation of 'sbuf' in the stack frame of function 'TranslateClause'
    #0 0x7f7f4423a1f0 in TranslateClause /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/translate.c:1941

While trying to match _NS1, MatchRule is overflowing the buffer.

It happens that this had not usually posed problem because rules usually
have these non-eating special characters last in the rule and thus it wasn't
mattering that post_ptr is pointing outside valid text.

When matching a double-letter whose encoding needs multiple bytes, to
properly skip the double-letter we have to skip these bytes, like other
rules.

Some rules test against character not being of a certain type. That may
match with the \0 beginning-of-text marker, and thus actually step over
it and let MatchRule continue with uninitialized data before it, leading
to potential random behavior.

This commits fixes it by making sure that we don't read before that \0.

LookupDict2 looks forward in the wtab array, it should still stop at its
end. Otherwise the memory sanitizer reports this:

testing en A. B C, D. E: F.
==65960==WARNING: MemorySanitizer: use-of-uninitialized-value
    #0 0x7ff9d7ef0de8 in LookupDict2 /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/dictionary.c:2676:11
    #1 0x7ff9d7eec2ec in LookupDictList /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/dictionary.c:2899:10
    #2 0x7ff9d802860a in TranslateWord3 /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/translate.c:588:12
    #3 0x7ff9d80249d4 in TranslateWord /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/translate.c:1100:14
    #4 0x7ff9d8051fe0 in TranslateWord2 /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/translate.c:1361:11
    #5 0x7ff9d804885c in TranslateClause /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/translate.c:2623:17
    #6 0x7ff9d800e4e9 in SpeakNextClause /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/synthesize.c:1569:2
    #7 0x7ff9d7fa50e6 in Synthesize /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/speech.c:457:2
    #8 0x7ff9d7fa41b3 in sync_espeak_Synth /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/speech.c:570:29
    #9 0x7ff9d7fa872f in espeak_ng_Synthesize /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/speech.c:678:10
    #10 0x7ff9d7f06584 in espeak_Synth /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/espeak_api.c:90:32
    #11 0x4a8be3 in main /home/samy/brl/speech/espeak-ng-git/src/espeak-ng.c:691:3
    #12 0x7ff9d78297fc in __libc_start_main csu/../csu/libc-start.c:332:16
    #13 0x421449 in _start (/home/samy/ens/projet/1/speech/espeak-ng-git/src/.libs/espeak-ng+0x421449)

  Uninitialized value was created by an allocation of 'words' in the stack frame of function 'TranslateClause'
    #0 0x7ff9d8035380 in TranslateClause /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/translate.c:1941

Strictly speaking, we are not supposed to use memcmp to compare strings
since we are not supposed to read beyond \0, which memcmp is supposed to
potentially do. Sanitizers would warn about it, and using strncmp happens to
provide the proper semantic while being not really slower, so better
just use them.

phonemes_name is only initialized when V_LANGUAGE is met. This is not
necessarily the case, notably with

testing espeak_SetVoiceByName("!v/Annie") (language variant; intonation)
Cannot set intonation: language not set, or is invalid.
Uninitialized bytes in __interceptor_strcmp at offset 0 inside [0x7fff8a875e30, 1)
==4169902==WARNING: MemorySanitizer: use-of-uninitialized-value
    #0 0x4c6a49 in LookupPhonemeTable /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/synthdata.c:363:7
    #1 0x4c6a49 in SelectPhonemeTableName /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/synthdata.c:380:12
    #2 0x5098a9 in LoadVoice /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/voices.c:950:34
    #3 0x50edcf in espeak_ng_SetVoiceByName /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/voices.c:1585:7
    #4 0x4aad63 in espeak_SetVoiceByName /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/espeak_api.c:125:32
    #5 0x4a3fe1 in test_espeak_set_voice_by_name_language_variant_intonation_parameter /home/samy/brl/speech/espeak-ng-git/tests/api.c:356:2
    #6 0x4a3fe1 in main /home/samy/brl/speech/espeak-ng-git/tests/api.c:567:2
    #7 0x7f26e88cb7fc in __libc_start_main csu/../csu/libc-start.c:332:16
    #8 0x4213a9 in _start (/home/samy/ens/projet/1/speech/espeak-ng-git/tests/api.test+0x4213a9)

  Uninitialized value was created by an allocation of 'phonemes_name' in the stack frame of function 'LoadVoice'
    #0 0x504290 in LoadVoice /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/voices.c:519

so better catch it properly rather than relying on uninitialized data.

The memory sanitizer would complain:

==4157154==WARNING: MemorySanitizer: use-of-uninitialized-value
    #0 0x7fc191d0a85b in TranslateWord3 /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/translate.c:1065:7
    #1 0x7fc191d02916 in TranslateWord /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/translate.c:1100:14
    #2 0x7fc191d1b324 in TranslateWord2 /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/translate.c:1448:15
    #3 0x7fc191d14ebc in TranslateClause /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/translate.c:2623:17
    #4 0x7fc191cfbc9b in SpeakNextClause /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/synthesize.c:1569:2
    #5 0x7fc191cd52fc in Synthesize /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/speech.c:457:2
    #6 0x7fc191cd6d7c in sync_espeak_Synth /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/speech.c:570:29
    #7 0x7fc191cd6d7c in espeak_ng_Synthesize /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/speech.c:678:10
    #8 0x7fc191ca0340 in espeak_Synth /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/espeak_api.c:90:32
    #9 0x4a4381 in main /home/samy/brl/speech/espeak-ng-git/src/espeak-ng.c:691:3
    #10 0x7fc19168b7fc in __libc_start_main csu/../csu/libc-start.c:332:16
    #11 0x421449 in _start (/home/samy/ens/projet/1/speech/espeak-ng-git/src/.libs/espeak-ng+0x421449)

  Uninitialized value was created by a heap allocation
    #0 0x45000d in malloc (/home/samy/ens/projet/1/speech/espeak-ng-git/src/.libs/espeak-ng+0x45000d)
    #1 0x7fc191d1ca29 in NewTranslator /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/tr_languages.c:242:26
    #2 0x7fc191d1ca29 in SelectTranslator /home/samy/brl/speech/espeak-ng-git/src/libespeak-ng/tr_languages.c:482:7

(and similar for expect_verb_sn expect_noun, expect_past,
clause_upper_count, clause_lower_count)

Indeed TranslateWord3 doesn't always initialize these fields. Better
just initialize them directly from the Translator creation.

The bw array is 0-based, not 1-based.

We need to keep the mutex around setting it, to make sure that the
visibility of setting it to true doesn't get delayed, notably not after
the signal (which could entail that the signal becomes useless, leading
to a hang).

speechPlayer_terminate was never called without calling
speechPlayer_initialize right after, so there would always be a FrameManager
laying behind.

strlen(p) may be arbitrarily long, that would underflow the word, for
instance:

testing fr Latn
=================================================================
==3741805==ERROR: AddressSanitizer: stack-buffer-overflow on address 0x7ffd733c1329 at pc 0x7ff5ffbad2de bp 0x7ffd733bf310 sp 0x7ffd733bf308
READ of size 1 at 0x7ffd733c1329 thread T0
    #0 0x7ff5ffbad2dd in IsLetterGroup src/libespeak-ng/dictionary.c:714
    #1 0x7ff5ffbbe425 in MatchRule src/libespeak-ng/dictionary.c:1979
    #2 0x7ff5ffbc09e9 in TranslateRules src/libespeak-ng/dictionary.c:2301
    #3 0x7ff5ffc26656 in TranslateWord3 src/libespeak-ng/translate.c:733
    #4 0x7ff5ffc2a10b in TranslateWord src/libespeak-ng/translate.c:1100
    #5 0x7ff5ffc2bef2 in TranslateWord2 src/libespeak-ng/translate.c:1361
    #6 0x7ff5ffc374e2 in TranslateClause src/libespeak-ng/translate.c:2623
    #7 0x7ff5ffc1d010 in SpeakNextClause src/libespeak-ng/synthesize.c:1569
    #8 0x7ff5ffbfbd46 in Synthesize src/libespeak-ng/speech.c:492
    #9 0x7ff5ffbfd52a in sync_espeak_Synth src/libespeak-ng/speech.c:570
    #10 0x7ff5ffbfdd1f in espeak_ng_Synthesize src/libespeak-ng/speech.c:678
    #11 0x7ff5ffbc72fd in espeak_Synth src/libespeak-ng/espeak_api.c:90
    #12 0x5627511a3137 in main src/espeak-ng.c:691
    #13 0x7ff5fee557fc in __libc_start_main ../csu/libc-start.c:332
    #14 0x5627511a0569 in _start (/home/samy/ens/projet/1/speech/espeak-ng-git/src/.libs/espeak-ng+0x6569)

Address 0x7ffd733c1329 is located in stack of thread T0 at offset 1177 in frame
    #0 0x7ff5ffc2f760 in TranslateClause src/libespeak-ng/translate.c:1941

  This frame has 16 object(s):
    [48, 52) 'cc' (line 1944)
    [64, 68) 'source_index' (line 1945)
    [80, 84) 'prev_in' (line 1948)
    [96, 100) 'prev_out' (line 1949)
    [112, 116) 'next_in' (line 1952)
    [128, 132) 'char_inserted' (line 1954)
    [144, 148) 'word_flags' (line 1963)
    [160, 164) 'charix_top' (line 1975)
    [176, 180) 'tone' (line 1985)
    [192, 196) 'next2_in' (line 2294)
    [208, 212) 'c_temp' (line 2518)
    [224, 374) 'number_buf' (line 2522)
    [448, 1048) 'num_wtab' (line 2523)
    [1184, 1984) 'sbuf' (line 1982) <== Memory access at offset 1177 underflows this variable
    [2112, 3720) 'charix' (line 1977)
    [3856, 7456) 'words' (line 1978)

sbuf is however properly '\0'-header, so we can make IsLetterGroup
carefully walk back in the word and issue a mismatch if it walks back
too much.

Fixes #1108

strlen(p) may be arbitrarily long, that would underflow the word, for
instance:

testing fr Latn
=================================================================
==3741805==ERROR: AddressSanitizer: stack-buffer-overflow on address 0x7ffd733c1329 at pc 0x7ff5ffbad2de bp 0x7ffd733bf310 sp 0x7ffd733bf308
READ of size 1 at 0x7ffd733c1329 thread T0
    #0 0x7ff5ffbad2dd in IsLetterGroup src/libespeak-ng/dictionary.c:714
    #1 0x7ff5ffbbe425 in MatchRule src/libespeak-ng/dictionary.c:1979
    #2 0x7ff5ffbc09e9 in TranslateRules src/libespeak-ng/dictionary.c:2301
    #3 0x7ff5ffc26656 in TranslateWord3 src/libespeak-ng/translate.c:733
    #4 0x7ff5ffc2a10b in TranslateWord src/libespeak-ng/translate.c:1100
    #5 0x7ff5ffc2bef2 in TranslateWord2 src/libespeak-ng/translate.c:1361
    #6 0x7ff5ffc374e2 in TranslateClause src/libespeak-ng/translate.c:2623
    #7 0x7ff5ffc1d010 in SpeakNextClause src/libespeak-ng/synthesize.c:1569
    #8 0x7ff5ffbfbd46 in Synthesize src/libespeak-ng/speech.c:492
    #9 0x7ff5ffbfd52a in sync_espeak_Synth src/libespeak-ng/speech.c:570
    #10 0x7ff5ffbfdd1f in espeak_ng_Synthesize src/libespeak-ng/speech.c:678
    #11 0x7ff5ffbc72fd in espeak_Synth src/libespeak-ng/espeak_api.c:90
    #12 0x5627511a3137 in main src/espeak-ng.c:691
    #13 0x7ff5fee557fc in __libc_start_main ../csu/libc-start.c:332
    #14 0x5627511a0569 in _start (/home/samy/ens/projet/1/speech/espeak-ng-git/src/.libs/espeak-ng+0x6569)

Address 0x7ffd733c1329 is located in stack of thread T0 at offset 1177 in frame
    #0 0x7ff5ffc2f760 in TranslateClause src/libespeak-ng/translate.c:1941

  This frame has 16 object(s):
    [48, 52) 'cc' (line 1944)
    [64, 68) 'source_index' (line 1945)
    [80, 84) 'prev_in' (line 1948)
    [96, 100) 'prev_out' (line 1949)
    [112, 116) 'next_in' (line 1952)
    [128, 132) 'char_inserted' (line 1954)
    [144, 148) 'word_flags' (line 1963)
    [160, 164) 'charix_top' (line 1975)
    [176, 180) 'tone' (line 1985)
    [192, 196) 'next2_in' (line 2294)
    [208, 212) 'c_temp' (line 2518)
    [224, 374) 'number_buf' (line 2522)
    [448, 1048) 'num_wtab' (line 2523)
    [1184, 1984) 'sbuf' (line 1982) <== Memory access at offset 1177 underflows this variable
    [2112, 3720) 'charix' (line 1977)
    [3856, 7456) 'words' (line 1978)

sbuf is however properly '\0'-header, so we can make IsLetterGroup
carefully walk back in the word and issue a mismatch if it walks back
too much.

Fixes #1108

TranslateWord2 uses phonemes in ph_list2. Apart from the breakable loops, it
may statically require up to 7 phonemes. Then TranslateClause always
uses 2 phonemes. We thus have to keep these margins along the loops to
avoid any overflow.

Fixes #1073 #1095

TranslateWord2 uses phonemes in ph_list2. Apart from the breakable loops, it
may statically require up to 7 phonemes. Then TranslateClause always
uses 2 phonemes. We thus have to keep these margins along the loops to
avoid any overflow.

Fixes #1073

Valgrind reports:

==3642987== Conditional jump or move depends on uninitialised value(s)
==3642987==    at 0x491F268: TranslateNumber_1 (numbers.c:1785)
==3642987==    by 0x4923C35: TranslateNumber (numbers.c:2080)
==3642987==    by 0x49556DC: TranslateWord3 (translate.c:644)
==3642987==    by 0x4957FCE: TranslateWord (translate.c:1100)
==3642987==    by 0x4959344: TranslateWord2 (translate.c:1361)
==3642987==    by 0x496116E: TranslateClause (translate.c:2613)
==3642987==    by 0x494FF7A: SpeakNextClause (synthesize.c:1569)
==3642987==    by 0x4939B9D: Synthesize (speech.c:457)
==3642987==    by 0x493AE6A: sync_espeak_Synth (speech.c:570)
==3642987==    by 0x493B286: espeak_ng_Synthesize (speech.c:678)
==3642987==    by 0x4916925: espeak_Synth (espeak_api.c:90)
==3642987==    by 0x10CF5D: main (espeak-ng.c:691)
==3642987==  Uninitialised value was created by a stack allocation
==3642987==    at 0x495BD9F: TranslateClause (translate.c:1941)

Indeed, TranslateNumber_1 looks back up to three bytes before, with
IsDigit09(word[-3])), so we have to increase the heading margin to three
spaces.

pre_ptr is already one byte before the current letter, so we do not want
to subtract 1 again. Otherwise this would for instance underflow word_iz
of addPluralSuffixes.

Otherwise asan reports this during make check:

testing en ibm mit ibms mits IBM MIT APH CES ITX IBMs MIT's APHs CES's ITXs
==3733154==ERROR: AddressSanitizer: stack-buffer-overflow on address 0x7ffe420233ef at pc 0x7f2e8a30aef1 bp 0x7ffe42022c80 sp 0x7ffe42022c78
READ of size 1 at 0x7ffe420233ef thread T0
    #0 0x7f2e8a30aef0 in utf8_in2 src/libespeak-ng/translate.c:281
    #1 0x7f2e8a2a6db1 in MatchRule src/libespeak-ng/dictionary.c:2058
    #2 0x7f2e8a2a89e9 in TranslateRules src/libespeak-ng/dictionary.c:2301
    #3 0x7f2e8a30cc77 in addPluralSuffixes src/libespeak-ng/translate.c:393
    #4 0x7f2e8a30e2c9 in TranslateWord3 src/libespeak-ng/translate.c:684
    #5 0x7f2e8a31210b in TranslateWord src/libespeak-ng/translate.c:1100
    #6 0x7f2e8a313ef2 in TranslateWord2 src/libespeak-ng/translate.c:1361
    #7 0x7f2e8a31f4e2 in TranslateClause src/libespeak-ng/translate.c:2623
    #8 0x7f2e8a305010 in SpeakNextClause src/libespeak-ng/synthesize.c:1569
    #9 0x7f2e8a2e390e in Synthesize src/libespeak-ng/speech.c:457
    #10 0x7f2e8a2e552a in sync_espeak_Synth src/libespeak-ng/speech.c:570
    #11 0x7f2e8a2e5d1f in espeak_ng_Synthesize src/libespeak-ng/speech.c:678
    #12 0x7f2e8a2af2fd in espeak_Synth src/libespeak-ng/espeak_api.c:90
    #13 0x5618104c9137 in main src/espeak-ng.c:691
    #14 0x7f2e8953d7fc in __libc_start_main ../csu/libc-start.c:332
    #15 0x5618104c6569 in _start (/home/samy/ens/projet/1/speech/espeak-ng-git/src/.libs/espeak-ng+0x6569)

Address 0x7ffe420233ef is located in stack of thread T0 at offset 47 in frame
    #0 0x7f2e8a30cb3b in addPluralSuffixes src/libespeak-ng/translate.c:380

  This frame has 3 object(s):
    [32, 36) 'word_zz' (line 381)
    [48, 52) 'word_iz' (line 382) <== Memory access at offset 47 underflows this variable
    [64, 68) 'word_ss' (line 383)

and indeed, RULE_NOVOWELS keeps looking back until it finds a spacing
character, so we have to provide it with one.

pre_ptr is already one byte before the current letter, so we do not want
to subtract 1 again. Otherwise this would for instance underflow word_iz
of addPluralSuffixes.

This avoids having to make LoadEnvelope2 pass &env_x[-1], which is very
strongly frowned upon by sanitizers such as usan.

TranslateWord2 uses phonemes in ph_list2. Apart from the breakable loops, it
may statically require up to 7 phonemes. Then TranslateClause always
uses 2 phonemes. We thus have to keep these margins along the loops to
avoid any overflow.

Fixes #1073

Valgrind reports:

==3642987== Conditional jump or move depends on uninitialised value(s)
==3642987==    at 0x491F268: TranslateNumber_1 (numbers.c:1785)
==3642987==    by 0x4923C35: TranslateNumber (numbers.c:2080)
==3642987==    by 0x49556DC: TranslateWord3 (translate.c:644)
==3642987==    by 0x4957FCE: TranslateWord (translate.c:1100)
==3642987==    by 0x4959344: TranslateWord2 (translate.c:1361)
==3642987==    by 0x496116E: TranslateClause (translate.c:2613)
==3642987==    by 0x494FF7A: SpeakNextClause (synthesize.c:1569)
==3642987==    by 0x4939B9D: Synthesize (speech.c:457)
==3642987==    by 0x493AE6A: sync_espeak_Synth (speech.c:570)
==3642987==    by 0x493B286: espeak_ng_Synthesize (speech.c:678)
==3642987==    by 0x4916925: espeak_Synth (espeak_api.c:90)
==3642987==    by 0x10CF5D: main (espeak-ng.c:691)
==3642987==  Uninitialised value was created by a stack allocation
==3642987==    at 0x495BD9F: TranslateClause (translate.c:1941)

Indeed, TranslateNumber_1 looks back up to three bytes before, with
IsDigit09(word[-3])), so we have to increase the heading margin to three
spaces.

valgrind reports

==3632264== Conditional jump or move depends on uninitialised value(s)
==3632264==    at 0x4846688: strcmp (vg_replace_strmem.c:924)
==3632264==    by 0x490EC12: LookupDictList (dictionary.c:2889)
==3632264==    by 0x49554C6: TranslateWord3 (translate.c:588)
==3632264==    by 0x4957FCE: TranslateWord (translate.c:1100)
==3632264==    by 0x4959344: TranslateWord2 (translate.c:1361)
==3632264==    by 0x4961390: TranslateClause (translate.c:2621)
==3632264==    by 0x494FF7A: SpeakNextClause (synthesize.c:1569)
==3632264==    by 0x4939B9D: Synthesize (speech.c:457)
==3632264==    by 0x493AE6A: sync_espeak_Synth (speech.c:570)
==3632264==    by 0x493B286: espeak_ng_Synthesize (speech.c:678)
==3632264==    by 0x4916925: espeak_Synth (espeak_api.c:90)
==3632264==    by 0x10CF5D: main (espeak-ng.c:691)

And indeed tr->phonemes_repeat may not necessarily be initialized.

shifting negative integers has an undefined behavior, so we have to
avoid it entirely. Using * 256 and / 256 provides the expected behavior
while being correctly optimized by the compiler.

event_notify currently introduces an arbitrary 50ms delay between speech
requests. This is usually unnoticed since it's small. But when
cancelling a long series of events, they add up to potentially seconds
of delays, while the user was precisely requesting to just cancel
everything as fast as possible.

This 50ms delay was probably meant to work around some issues elsewhere.
If they are still there, they should be fixed, not worked around.

https://github.com/espeak-ng/espeak-ng/issues/1072

Signed-off-by: Ulrich Müller <[email protected]>

envelope_dat.

envelope_dat and envelope_paths were only set but never used.
LoadEnvelope() and LoadEnvelope2() no longer need fname as an argument.

This way we can compile multiple dictionaries at a time without
stepping on 'temp' file from different compilation steps:

    $ make -j8 -B
    ...
    make[1]: *** [Makefile:3082: espeak-ng-data/an_dict] Segmentation fault (core dumped)
    make[1]: *** Deleting file 'espeak-ng-data/an_dict'
    make[1]: *** [Makefile:3082: espeak-ng-data/az_dict] Segmentation fault (core dumped)
    make[1]: *** Deleting file 'espeak-ng-data/az_dict'

Otherwise espeak_SetVoiceByName returns EE_OK even if the mbrola voice
file could not be loaded and thus setting the voice actually failed.

This will make it easier to set a default voice other than
English. This is important for cases when a language will fall back to
the default voice.

Some references to L('e', 'n') still need to be changed.

Found by parsing the results from this bash script:
for define in $(grep define *.h |  awk'{print $2}');
do if ! grep -q $define *.c
then
        echo "$define not found";
fi
done

This is unnecessary since the string is already null terminated in the
V_LANGUAGE switch case. sscanf creates a null terminated string in
language_name, which is then copied to voice_language by strcpy (keeping
the \0).

It's probably unnecessary code and could be removed. However, it would
break tests. This is a temporary solution.

The solution is to do it once in VoiceReset() and then if the "formant"
keyword changes voice->width.