From eb24399a17b85fad292fe5137bb9ea641f8b7896 Mon Sep 17 00:00:00 2001 From: dkf Date: Sun, 23 Feb 2014 13:07:36 +0000 Subject: [3597178]: Improve documentation of what's going on with encodings --- doc/encoding.n | 34 +++++++++++++++++++++++++++------- doc/string.n | 28 ++++++++++++++++++++++++---- 2 files changed, 51 insertions(+), 11 deletions(-) diff --git a/doc/encoding.n b/doc/encoding.n index be1dc3f..5782199 100644 --- a/doc/encoding.n +++ b/doc/encoding.n @@ -14,10 +14,21 @@ encoding \- Manipulate encodings .BE .SH INTRODUCTION .PP -Strings in Tcl are encoded using 16-bit Unicode characters. Different -operating system interfaces or applications may generate strings in -other encodings such as Shift-JIS. The \fBencoding\fR command helps -to bridge the gap between Unicode and these other formats. +Strings in Tcl are logically a sequence of 16-bit Unicode characters. +These strings are represented in memory as a sequence of bytes that +may be in one of several encodings: modified UTF\-8 (which uses 1 to 3 +bytes per character), 16-bit +.QW Unicode +(which uses 2 bytes per character, with an endianness that is +dependent on the host architecture), and binary (which uses a single +byte per character but only handles a restricted range of characters). +Tcl does not guarantee to always use the same encoding for the same +string. +.PP +Different operating system interfaces or applications may generate +strings in other encodings such as Shift\-JIS. The \fBencoding\fR +command helps to bridge the gap between Unicode and these other +formats. .SH DESCRIPTION .PP Performs one of several encoding related operations, depending on @@ -37,8 +48,9 @@ system encoding is used. Convert \fIstring\fR from Unicode to the specified \fIencoding\fR. The result is a sequence of bytes that represents the converted string. Each byte is stored in the lower 8-bits of a Unicode -character. If \fIencoding\fR is not specified, the current -system encoding is used. +character (indeed, the resulting string is a binary string as far as +Tcl is concerned, at least initially). If \fIencoding\fR is not +specified, the current system encoding is used. .TP \fBencoding dirs\fR ?\fIdirectoryList\fR? . @@ -56,6 +68,11 @@ searchable directory, that element is ignored. . Returns a list containing the names of all of the encodings that are currently available. +The encodings +.QW utf-8 +and +.QW iso8859-1 +are guaranteed to be present in the list. .TP \fBencoding system\fR ?\fIencoding\fR? . @@ -73,7 +90,7 @@ However, because the \fBsource\fR command always reads files using the current system encoding, Tcl will only source such files correctly when the encoding used to write the file is the same. This tends not to be true in an internationalized setting. For example, if such a -file was sourced in North America (where the ISO8859-1 is normally +file was sourced in North America (where the ISO8859\-1 is normally used), each byte in the file would be treated as a separate character that maps to the 00 page in Unicode. The resulting Tcl strings will not contain the expected Japanese characters. Instead, they will @@ -93,3 +110,6 @@ which is the Hiragana letter HA. Tcl_GetEncoding(3) .SH KEYWORDS encoding, unicode +.\" Local Variables: +.\" mode: nroff +.\" End: diff --git a/doc/string.n b/doc/string.n index 76005fc..163abdd 100644 --- a/doc/string.n +++ b/doc/string.n @@ -343,10 +343,13 @@ misleading. \fBstring bytelength \fIstring\fR . Returns a decimal string giving the number of bytes used to represent -\fIstring\fR in memory. Because UTF\-8 uses one to three bytes to -represent Unicode characters, the byte length will not be the same as -the character length in general. The cases where a script cares about -the byte length are rare. +\fIstring\fR in memory when encoded as Tcl's internal modified UTF\-8; +Tcl may use other encodings for \fIstring\fR as well, and does not +guarantee to only use a single encoding for a particular \fIstring\fR. +Because UTF\-8 uses a variable number of bytes to represent Unicode +characters, the byte length will not be the same as the character +length in general. The cases where a script cares about the byte +length are rare. .RS .PP In almost all cases, you should use the @@ -354,10 +357,27 @@ In almost all cases, you should use the Tcl byte array value). Refer to the \fBTcl_NumUtfChars\fR manual entry for more details on the UTF\-8 representation. .PP +Formally, the \fBstring bytelength\fR operation returns the content of +the \fIlength\fR field of the \fBTcl_Obj\fR structure, after calling +\fBTcl_GetString\fR to ensure that the \fIbytes\fR field is populated. +This is highly unlikely to be useful to Tcl scripts, as Tcl's internal +encoding is not strict UTF\-8, but rather a modified CESU\-8 with a +denormalized NUL (identical to that used in a number of places by +Java's serialization mechanism) to enable basic processing with +non-Unicode-aware C functions. As this representation should only +ever be used by Tcl's implementation, the number of bytes used to +store the representation is of very low value (except to C extension +code, which has direct access for the purpose of memory management, +etc.) +.PP \fICompatibility note:\fR it is likely that this subcommand will be withdrawn in a future version of Tcl. It is better to use the \fBencoding convertto\fR command to convert a string to a known encoding and then apply \fBstring length\fR to that. +.PP +.CS +\fBstring length\fR [encoding convertto utf-8 $theString] +.CE .RE .TP \fBstring wordend \fIstring charIndex\fR -- cgit v0.12