diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-07 08:50:31 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-07 08:50:31 +0000 |
commit | aed8ce9da277f5ecffe968b324f242c41c3b752a (patch) | |
tree | d2e538394cb7a8a7c42a4aac6ccf1a8e3256999b /src/winclip.c | |
parent | Initial commit. (diff) | |
download | vim-aed8ce9da277f5ecffe968b324f242c41c3b752a.tar.xz vim-aed8ce9da277f5ecffe968b324f242c41c3b752a.zip |
Adding upstream version 2:9.0.1378.upstream/2%9.0.1378upstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'src/winclip.c')
-rw-r--r-- | src/winclip.c | 774 |
1 files changed, 774 insertions, 0 deletions
diff --git a/src/winclip.c b/src/winclip.c new file mode 100644 index 0000000..8f54c6b --- /dev/null +++ b/src/winclip.c @@ -0,0 +1,774 @@ +/* vi:set ts=8 sts=4 sw=4 noet: + * + * 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. + */ + +/* + * winclip.c + * + * Routines for Win32 clipboard handling. + * Also used by Cygwin, using os_unix.c. + */ + +#include "vim.h" + +/* + * Compile only the clipboard handling features when compiling for cygwin + * posix environment. + */ +#ifdef FEAT_CYGWIN_WIN32_CLIPBOARD +# define WIN32_LEAN_AND_MEAN +# include <windows.h> +# include "winclip.pro" +#endif + +/* + * When generating prototypes for Win32 on Unix, these lines make the syntax + * errors disappear. They do not need to be correct. + */ +#ifdef PROTO +#define WINAPI +#define WINBASEAPI +typedef int DWORD; +typedef int LPBOOL; +typedef int LPCSTR; +typedef int LPCWSTR; +typedef int LPSTR; +typedef int LPWSTR; +typedef int UINT; +#endif + +/* + * Convert an UTF-8 string to UTF-16. + * "instr[inlen]" is the input. "inlen" is in bytes. + * When "outstr" is NULL only return the number of UTF-16 words produced. + * Otherwise "outstr" must be a buffer of sufficient size. + * Returns the number of UTF-16 words produced. + */ + int +utf8_to_utf16(char_u *instr, int inlen, short_u *outstr, int *unconvlenp) +{ + int outlen = 0; + char_u *p = instr; + int todo = inlen; + int l; + int ch; + + while (todo > 0) + { + // Only convert if we have a complete sequence. + l = utf_ptr2len_len(p, todo); + if (l > todo) + { + // Return length of incomplete sequence. + if (unconvlenp != NULL) + *unconvlenp = todo; + break; + } + + ch = utf_ptr2char(p); + if (ch >= 0x10000) + { + // non-BMP character, encoding with surrogate pairs + ++outlen; + if (outstr != NULL) + { + *outstr++ = (0xD800 - (0x10000 >> 10)) + (ch >> 10); + *outstr++ = 0xDC00 | (ch & 0x3FF); + } + } + else if (outstr != NULL) + *outstr++ = ch; + ++outlen; + p += l; + todo -= l; + } + + return outlen; +} + +/* + * Convert an UTF-16 string to UTF-8. + * The input is "instr[inlen]" with "inlen" in number of UTF-16 words. + * When "outstr" is NULL only return the required number of bytes. + * Otherwise "outstr" must be a buffer of sufficient size. + * Return the number of bytes produced. + */ + int +utf16_to_utf8(short_u *instr, int inlen, char_u *outstr) +{ + int outlen = 0; + int todo = inlen; + short_u *p = instr; + int l; + int ch, ch2; + + while (todo > 0) + { + ch = *p; + if (ch >= 0xD800 && ch <= 0xDBFF && todo > 1) + { + // surrogate pairs handling + ch2 = p[1]; + if (ch2 >= 0xDC00 && ch2 <= 0xDFFF) + { + ch = ((ch - 0xD800) << 10) + (ch2 & 0x3FF) + 0x10000; + ++p; + --todo; + } + } + if (outstr != NULL) + { + l = utf_char2bytes(ch, outstr); + outstr += l; + } + else + l = utf_char2len(ch); + ++p; + outlen += l; + --todo; + } + + return outlen; +} + +/* + * Call MultiByteToWideChar() and allocate memory for the result. + * Returns the result in "*out[*outlen]" with an extra zero appended. + * "outlen" is in words. + */ + void +MultiByteToWideChar_alloc(UINT cp, DWORD flags, + LPCSTR in, int inlen, + LPWSTR *out, int *outlen) +{ + *outlen = MultiByteToWideChar(cp, flags, in, inlen, 0, 0); + // Add one word to avoid a zero-length alloc(). + *out = ALLOC_MULT(WCHAR, *outlen + 1); + if (*out == NULL) + return; + + MultiByteToWideChar(cp, flags, in, inlen, *out, *outlen); + (*out)[*outlen] = 0; +} + +/* + * Call WideCharToMultiByte() and allocate memory for the result. + * Returns the result in "*out[*outlen]" with an extra NUL appended. + */ + void +WideCharToMultiByte_alloc(UINT cp, DWORD flags, + LPCWSTR in, int inlen, + LPSTR *out, int *outlen, + LPCSTR def, LPBOOL useddef) +{ + *outlen = WideCharToMultiByte(cp, flags, in, inlen, NULL, 0, def, useddef); + // Add one byte to avoid a zero-length alloc(). + *out = alloc(*outlen + 1); + if (*out == NULL) + return; + WideCharToMultiByte(cp, flags, in, inlen, *out, *outlen, def, useddef); + (*out)[*outlen] = 0; +} + + +#ifdef FEAT_CLIPBOARD +/* + * Clipboard stuff, for cutting and pasting text to other windows. + */ + + void +win_clip_init(void) +{ + clip_init(TRUE); + + /* + * Vim's own clipboard format recognises whether the text is char, line, + * or rectangular block. Only useful for copying between two Vims. + * "Clipboard_T" was used for previous versions, using the first + * character to specify MCHAR, MLINE or MBLOCK. + */ + clip_star.format = RegisterClipboardFormat("VimClipboard2"); + clip_star.format_raw = RegisterClipboardFormat("VimRawBytes"); +} + +// Type used for the clipboard type of Vim's data. +typedef struct +{ + int type; // MCHAR, MBLOCK or MLINE + int txtlen; // length of CF_TEXT in bytes + int ucslen; // length of CF_UNICODETEXT in words + int rawlen; // length of clip_star.format_raw, including encoding, + // excluding terminating NUL +} VimClipType_t; + +/* + * Make vim the owner of the current selection. Return OK upon success. + */ + int +clip_mch_own_selection(Clipboard_T *cbd UNUSED) +{ + /* + * Never actually own the clipboard. If another application sets the + * clipboard, we don't want to think that we still own it. + */ + return FAIL; +} + +/* + * Make vim NOT the owner of the current selection. + */ + void +clip_mch_lose_selection(Clipboard_T *cbd UNUSED) +{ + // Nothing needs to be done here +} + +/* + * Copy "str[*size]" into allocated memory, changing CR-NL to NL. + * Return the allocated result and the size in "*size". + * Returns NULL when out of memory. + */ + static char_u * +crnl_to_nl(const char_u *str, int *size) +{ + int pos = 0; + int str_len = *size; + char_u *ret; + char_u *retp; + + // Avoid allocating zero bytes, it generates an error message. + ret = alloc(str_len == 0 ? 1 : str_len); + if (ret != NULL) + { + retp = ret; + for (pos = 0; pos < str_len; ++pos) + { + if (str[pos] == '\r' && str[pos + 1] == '\n') + { + ++pos; + --(*size); + } + *retp++ = str[pos]; + } + } + + return ret; +} + +/* + * Wait for another process to Close the Clipboard. + * Returns TRUE for success. + */ + static int +vim_open_clipboard(void) +{ + int delay = 10; + + while (!OpenClipboard(NULL)) + { + if (delay > 500) + return FALSE; // waited too long, give up + Sleep(delay); + delay *= 2; // wait for 10, 20, 40, 80, etc. msec + } + return TRUE; +} + +/* + * Get the current selection and put it in the clipboard register. + * + * NOTE: Must use GlobalLock/Unlock here to ensure Win32s compatibility. + * On NT/W95 the clipboard data is a fixed global memory object and + * so its handle = its pointer. + * On Win32s, however, co-operation with the Win16 system means that + * the clipboard data is moveable and its handle is not a pointer at all, + * so we can't just cast the return value of GetClipboardData to (char_u*). + * <VN> + */ + void +clip_mch_request_selection(Clipboard_T *cbd) +{ + VimClipType_t metadata = { -1, -1, -1, -1 }; + HGLOBAL hMem = NULL; + char_u *str = NULL; + char_u *to_free = NULL; + HGLOBAL rawh = NULL; + int str_size = 0; + int maxlen; + size_t n; + + /* + * Don't pass GetActiveWindow() as an argument to OpenClipboard() because + * then we can't paste back into the same window for some reason - webb. + */ + if (!vim_open_clipboard()) + return; + + // Check for vim's own clipboard format first. This only gets the type of + // the data, still need to use CF_UNICODETEXT or CF_TEXT for the text. + if (IsClipboardFormatAvailable(cbd->format)) + { + VimClipType_t *meta_p; + HGLOBAL meta_h; + + // We have metadata on the clipboard; try to get it. + if ((meta_h = GetClipboardData(cbd->format)) != NULL + && (meta_p = (VimClipType_t *)GlobalLock(meta_h)) != NULL) + { + // The size of "VimClipType_t" changed, "rawlen" was added later. + // Only copy what is available for backwards compatibility. + n = sizeof(VimClipType_t); + if (GlobalSize(meta_h) < n) + n = GlobalSize(meta_h); + memcpy(&metadata, meta_p, n); + GlobalUnlock(meta_h); + } + } + + // Check for Vim's raw clipboard format first. This is used without + // conversion, but only if 'encoding' matches. + if (IsClipboardFormatAvailable(cbd->format_raw) + && metadata.rawlen > (int)STRLEN(p_enc)) + { + // We have raw data on the clipboard; try to get it. + if ((rawh = GetClipboardData(cbd->format_raw)) != NULL) + { + char_u *rawp; + + rawp = (char_u *)GlobalLock(rawh); + if (rawp != NULL && STRCMP(p_enc, rawp) == 0) + { + n = STRLEN(p_enc) + 1; + str = rawp + n; + str_size = (int)(metadata.rawlen - n); + } + else + { + GlobalUnlock(rawh); + rawh = NULL; + } + } + } + if (str == NULL) + { + // Try to get the clipboard in Unicode if it's not an empty string. + if (IsClipboardFormatAvailable(CF_UNICODETEXT) && metadata.ucslen != 0) + { + HGLOBAL hMemW; + + if ((hMemW = GetClipboardData(CF_UNICODETEXT)) != NULL) + { + WCHAR *hMemWstr = (WCHAR *)GlobalLock(hMemW); + + // Use the length of our metadata if possible, but limit it to + // the GlobalSize() for safety. + maxlen = (int)(GlobalSize(hMemW) / sizeof(WCHAR)); + if (metadata.ucslen >= 0) + { + if (metadata.ucslen > maxlen) + str_size = maxlen; + else + str_size = metadata.ucslen; + } + else + { + for (str_size = 0; str_size < maxlen; ++str_size) + if (hMemWstr[str_size] == NUL) + break; + } + to_free = str = utf16_to_enc((short_u *)hMemWstr, &str_size); + GlobalUnlock(hMemW); + } + } + // Get the clipboard in the Active codepage. + else if (IsClipboardFormatAvailable(CF_TEXT)) + { + if ((hMem = GetClipboardData(CF_TEXT)) != NULL) + { + str = (char_u *)GlobalLock(hMem); + + // The length is either what our metadata says or the strlen(). + // But limit it to the GlobalSize() for safety. + maxlen = (int)GlobalSize(hMem); + if (metadata.txtlen >= 0) + { + if (metadata.txtlen > maxlen) + str_size = maxlen; + else + str_size = metadata.txtlen; + } + else + { + for (str_size = 0; str_size < maxlen; ++str_size) + if (str[str_size] == NUL) + break; + } + + // The text is in the active codepage. Convert to + // 'encoding', going through UTF-16. + acp_to_enc(str, str_size, &to_free, &maxlen); + if (to_free != NULL) + { + str_size = maxlen; + str = to_free; + } + } + } + } + + if (str != NULL && metadata.txtlen != 0) + { + char_u *temp_clipboard; + + // If the type is not known detect it. + if (metadata.type == -1) + metadata.type = MAUTO; + + // Translate <CR><NL> into <NL>. + temp_clipboard = crnl_to_nl(str, &str_size); + if (temp_clipboard != NULL) + { + clip_yank_selection(metadata.type, temp_clipboard, str_size, cbd); + vim_free(temp_clipboard); + } + } + + // unlock the global object + if (hMem != NULL) + GlobalUnlock(hMem); + if (rawh != NULL) + GlobalUnlock(rawh); + CloseClipboard(); + vim_free(to_free); +} + +/* + * Send the current selection to the clipboard. + */ + void +clip_mch_set_selection(Clipboard_T *cbd) +{ + char_u *str = NULL; + VimClipType_t metadata; + long_u txtlen; + HGLOBAL hMemRaw = NULL; + HGLOBAL hMem = NULL; + HGLOBAL hMemVim = NULL; + HGLOBAL hMemW = NULL; + + // If the '*' register isn't already filled in, fill it in now + cbd->owned = TRUE; + clip_get_selection(cbd); + cbd->owned = FALSE; + + // Get the text to be put on the clipboard, with CR-LF. + metadata.type = clip_convert_selection(&str, &txtlen, cbd); + if (metadata.type < 0) + return; + metadata.txtlen = (int)txtlen; + metadata.ucslen = 0; + metadata.rawlen = 0; + + // Always set the raw bytes: 'encoding', NUL and the text. This is used + // when copy/paste from/to Vim with the same 'encoding', so that illegal + // bytes can also be copied and no conversion is needed. + { + LPSTR lpszMemRaw; + + metadata.rawlen = (int)(txtlen + STRLEN(p_enc) + 1); + hMemRaw = (LPSTR)GlobalAlloc(GMEM_MOVEABLE | GMEM_DDESHARE, + metadata.rawlen + 1); + lpszMemRaw = (LPSTR)GlobalLock(hMemRaw); + if (lpszMemRaw != NULL) + { + STRCPY(lpszMemRaw, p_enc); + memcpy(lpszMemRaw + STRLEN(p_enc) + 1, str, txtlen + 1); + GlobalUnlock(hMemRaw); + } + else + metadata.rawlen = 0; + } + + { + WCHAR *out; + int len = metadata.txtlen; + + // Convert the text to UTF-16. This is put on the clipboard as + // CF_UNICODETEXT. + out = (WCHAR *)enc_to_utf16(str, &len); + if (out != NULL) + { + WCHAR *lpszMemW; + + // Convert the text for CF_TEXT to Active codepage. Otherwise it's + // p_enc, which has no relation to the Active codepage. + metadata.txtlen = WideCharToMultiByte(GetACP(), 0, out, len, + NULL, 0, 0, 0); + vim_free(str); + str = alloc(metadata.txtlen == 0 ? 1 : metadata.txtlen); + if (str == NULL) + { + vim_free(out); + return; // out of memory + } + WideCharToMultiByte(GetACP(), 0, out, len, + (LPSTR)str, metadata.txtlen, 0, 0); + + // Allocate memory for the UTF-16 text, add one NUL word to + // terminate the string. + hMemW = (LPSTR)GlobalAlloc(GMEM_MOVEABLE | GMEM_DDESHARE, + (len + 1) * sizeof(WCHAR)); + lpszMemW = (WCHAR *)GlobalLock(hMemW); + if (lpszMemW != NULL) + { + memcpy(lpszMemW, out, len * sizeof(WCHAR)); + lpszMemW[len] = NUL; + GlobalUnlock(hMemW); + } + vim_free(out); + metadata.ucslen = len; + } + } + + // Allocate memory for the text, add one NUL byte to terminate the string. + hMem = GlobalAlloc(GMEM_MOVEABLE | GMEM_DDESHARE, metadata.txtlen + 1); + { + LPSTR lpszMem = (LPSTR)GlobalLock(hMem); + + if (lpszMem) + { + mch_memmove((char_u *)lpszMem, str, metadata.txtlen); + GlobalUnlock(hMem); + } + } + + // Set up metadata: + { + VimClipType_t *lpszMemVim = NULL; + + hMemVim = GlobalAlloc(GMEM_MOVEABLE|GMEM_DDESHARE, + sizeof(VimClipType_t)); + lpszMemVim = (VimClipType_t *)GlobalLock(hMemVim); + memcpy(lpszMemVim, &metadata, sizeof(metadata)); + GlobalUnlock(hMemVim); + } + + /* + * Open the clipboard, clear it and put our text on it. + * Always set our Vim format. Put Unicode and plain text on it. + * + * Don't pass GetActiveWindow() as an argument to OpenClipboard() + * because then we can't paste back into the same window for some + * reason - webb. + */ + if (vim_open_clipboard()) + { + if (EmptyClipboard()) + { + SetClipboardData(cbd->format, hMemVim); + hMemVim = 0; + if (hMemW != NULL) + { + if (SetClipboardData(CF_UNICODETEXT, hMemW) != NULL) + hMemW = NULL; + } + // Always use CF_TEXT. On Win98 Notepad won't obtain the + // CF_UNICODETEXT text, only CF_TEXT. + SetClipboardData(CF_TEXT, hMem); + hMem = 0; + } + CloseClipboard(); + } + + vim_free(str); + // Free any allocations we didn't give to the clipboard: + if (hMemRaw) + GlobalFree(hMemRaw); + if (hMem) + GlobalFree(hMem); + if (hMemW) + GlobalFree(hMemW); + if (hMemVim) + GlobalFree(hMemVim); +} + +#endif // FEAT_CLIPBOARD + +/* + * Note: the following two functions are only guaranteed to work when using + * valid MS-Windows codepages or when iconv() is available. + */ + +/* + * Convert "str" from 'encoding' to UTF-16. + * Input in "str" with length "*lenp". When "lenp" is NULL, use strlen(). + * Output is returned as an allocated string. "*lenp" is set to the length of + * the result. A trailing NUL is always added. + * Returns NULL when out of memory. + */ + short_u * +enc_to_utf16(char_u *str, int *lenp) +{ + vimconv_T conv; + WCHAR *ret; + char_u *allocbuf = NULL; + int len_loc; + int length; + + if (lenp == NULL) + { + len_loc = (int)STRLEN(str) + 1; + lenp = &len_loc; + } + + if (enc_codepage > 0) + { + // We can do any CP### -> UTF-16 in one pass, and we can do it + // without iconv() (convert_* may need iconv). + MultiByteToWideChar_alloc(enc_codepage, 0, (LPCSTR)str, *lenp, + &ret, &length); + } + else + { + // Use "latin1" by default, we might be called before we have p_enc + // set up. Convert to utf-8 first, works better with iconv(). Does + // nothing if 'encoding' is "utf-8". + conv.vc_type = CONV_NONE; + if (convert_setup(&conv, p_enc ? p_enc : (char_u *)"latin1", + (char_u *)"utf-8") == FAIL) + return NULL; + if (conv.vc_type != CONV_NONE) + { + str = allocbuf = string_convert(&conv, str, lenp); + if (str == NULL) + return NULL; + } + convert_setup(&conv, NULL, NULL); + + length = utf8_to_utf16(str, *lenp, NULL, NULL); + ret = ALLOC_MULT(WCHAR, length + 1); + if (ret != NULL) + { + utf8_to_utf16(str, *lenp, (short_u *)ret, NULL); + ret[length] = 0; + } + + vim_free(allocbuf); + } + + *lenp = length; + return (short_u *)ret; +} + +/* + * Convert an UTF-16 string to 'encoding'. + * Input in "str" with length (counted in wide characters) "*lenp". When + * "lenp" is NULL, use wcslen(). + * Output is returned as an allocated string. If "*lenp" is not NULL it is + * set to the length of the result. + * Returns NULL when out of memory. + */ + char_u * +utf16_to_enc(short_u *str, int *lenp) +{ + vimconv_T conv; + char_u *utf8_str = NULL, *enc_str = NULL; + int len_loc; + + if (lenp == NULL) + { + len_loc = (int)wcslen(str) + 1; + lenp = &len_loc; + } + + if (enc_codepage > 0) + { + // We can do any UTF-16 -> CP### in one pass. + int length; + + WideCharToMultiByte_alloc(enc_codepage, 0, str, *lenp, + (LPSTR *)&enc_str, &length, 0, 0); + *lenp = length; + return enc_str; + } + + // Avoid allocating zero bytes, it generates an error message. + utf8_str = alloc(utf16_to_utf8(str, *lenp == 0 ? 1 : *lenp, NULL)); + if (utf8_str != NULL) + { + *lenp = utf16_to_utf8(str, *lenp, utf8_str); + + // We might be called before we have p_enc set up. + conv.vc_type = CONV_NONE; + convert_setup(&conv, (char_u *)"utf-8", + p_enc? p_enc: (char_u *)"latin1"); + if (conv.vc_type == CONV_NONE) + { + // p_enc is utf-8, so we're done. + enc_str = utf8_str; + } + else + { + enc_str = string_convert(&conv, utf8_str, lenp); + vim_free(utf8_str); + } + + convert_setup(&conv, NULL, NULL); + } + + return enc_str; +} + +/* + * Convert from the active codepage to 'encoding'. + * Input is "str[str_size]". + * The result is in allocated memory: "out[outlen]". "outlen" includes the + * terminating NUL. + */ + void +acp_to_enc( + char_u *str, + int str_size, + char_u **out, + int *outlen) + +{ + LPWSTR widestr; + + MultiByteToWideChar_alloc(GetACP(), 0, (LPCSTR)str, str_size, + &widestr, outlen); + if (widestr == NULL) + return; + ++*outlen; // Include the 0 after the string + *out = utf16_to_enc((short_u *)widestr, outlen); + vim_free(widestr); +} + +/* + * Convert from 'encoding' to the active codepage. + * Input is "str[str_size]". + * The result is in allocated memory: "out[outlen]". With terminating NUL. + */ + void +enc_to_acp( + char_u *str, + int str_size, + char_u **out, + int *outlen) + +{ + LPWSTR widestr; + int len = str_size; + + widestr = (WCHAR *)enc_to_utf16(str, &len); + if (widestr == NULL) + return; + WideCharToMultiByte_alloc(GetACP(), 0, widestr, len, + (LPSTR *)out, outlen, 0, 0); + vim_free(widestr); +} |