static svn_error_t *
read_handler_apr(void *baton, char *buffer, apr_size_t *len)
{
struct baton_apr *btn = baton;
svn_error_t *err;
svn_boolean_t eof;
if (*len == 1)
{
err = svn_io_file_getc(buffer, btn->file, btn->pool);
if (err)
{
*len = 0;
if (APR_STATUS_IS_EOF(err->apr_err))
{
svn_error_clear(err);
err = SVN_NO_ERROR;
}
}
}
else
err = svn_io_file_read_full2(btn->file, buffer, *len, len,
&eof, btn->pool);
return svn_error_trace(err);
}
svn_error_t *
svn_fs_x__rev_file_read(svn_fs_x__revision_file_t *file,
void *buf,
apr_size_t nbytes)
{
SVN_ERR(auto_open(file));
return svn_error_trace(svn_io_file_read_full2(file->file, buf, nbytes,
NULL, NULL, file->pool));
}
/* If the footer data in FILE has not been read, yet, do so now.
* Index locations will only be read upon request as we assume they get
* cached and the FILE is usually used for REP data access only.
* Hence, the separate step.
*/
static svn_error_t *
auto_read_footer(svn_fs_x__revision_file_t *file)
{
if (file->l2p_info.start == -1)
{
apr_off_t filesize = 0;
unsigned char footer_length;
svn_stringbuf_t *footer;
/* Determine file size. */
SVN_ERR(auto_open(file));
SVN_ERR(svn_io_file_seek(file->file, APR_END, &filesize, file->pool));
/* Read last byte (containing the length of the footer). */
SVN_ERR(svn_io_file_aligned_seek(file->file, file->block_size, NULL,
filesize - 1, file->pool));
SVN_ERR(svn_io_file_read_full2(file->file, &footer_length,
sizeof(footer_length), NULL, NULL,
file->pool));
/* Read footer. */
footer = svn_stringbuf_create_ensure(footer_length, file->pool);
SVN_ERR(svn_io_file_aligned_seek(file->file, file->block_size, NULL,
filesize - 1 - footer_length,
file->pool));
SVN_ERR(svn_io_file_read_full2(file->file, footer->data, footer_length,
&footer->len, NULL, file->pool));
footer->data[footer->len] = '\0';
/* Extract index locations. */
SVN_ERR(svn_fs_x__parse_footer(&file->l2p_info.start,
&file->l2p_info.checksum,
&file->p2l_info.start,
&file->p2l_info.checksum,
footer, file->file_info.start_revision,
filesize - footer_length - 1, file->pool));
file->l2p_info.end = file->p2l_info.start;
file->p2l_info.end = filesize - footer_length - 1;
}
return SVN_NO_ERROR;
}
/* There are enough quirks in the deprecated svn_hash_read that we
should just preserve its implementation. */
svn_error_t *
svn_hash_read(apr_hash_t *hash,
apr_file_t *srcfile,
apr_pool_t *pool)
{
svn_error_t *err;
char buf[SVN_KEYLINE_MAXLEN];
apr_size_t num_read;
char c;
int first_time = 1;
while (1)
{
/* Read a key length line. Might be END, though. */
apr_size_t len = sizeof(buf);
err = svn_io_read_length_line(srcfile, buf, &len, pool);
if (err && APR_STATUS_IS_EOF(err->apr_err) && first_time)
{
/* We got an EOF on our very first attempt to read, which
means it's a zero-byte file. No problem, just go home. */
svn_error_clear(err);
return SVN_NO_ERROR;
}
else if (err)
/* Any other circumstance is a genuine error. */
return err;
first_time = 0;
if (((len == 3) && (buf[0] == 'E') && (buf[1] == 'N') && (buf[2] == 'D'))
|| ((len == 9)
&& (buf[0] == 'P')
&& (buf[1] == 'R') /* We formerly used just "END" to */
&& (buf[2] == 'O') /* end a property hash, but later */
&& (buf[3] == 'P') /* we added "PROPS-END", so that */
&& (buf[4] == 'S') /* the fs dump format would be */
&& (buf[5] == '-') /* more human-readable. That's */
&& (buf[6] == 'E') /* why we accept either way here. */
&& (buf[7] == 'N')
&& (buf[8] == 'D')))
{
/* We've reached the end of the dumped hash table, so leave. */
return SVN_NO_ERROR;
}
else if ((buf[0] == 'K') && (buf[1] == ' '))
{
size_t keylen;
int parsed_len;
void *keybuf;
/* Get the length of the key */
SVN_ERR(svn_cstring_atoi(&parsed_len, buf + 2));
keylen = parsed_len;
/* Now read that much into a buffer, + 1 byte for null terminator */
keybuf = apr_palloc(pool, keylen + 1);
SVN_ERR(svn_io_file_read_full2(srcfile,
keybuf, keylen,
&num_read, NULL, pool));
((char *) keybuf)[keylen] = '\0';
/* Suck up extra newline after key data */
SVN_ERR(svn_io_file_getc(&c, srcfile, pool));
if (c != '\n')
return svn_error_create(SVN_ERR_MALFORMED_FILE, NULL, NULL);
/* Read a val length line */
len = sizeof(buf);
SVN_ERR(svn_io_read_length_line(srcfile, buf, &len, pool));
if ((buf[0] == 'V') && (buf[1] == ' '))
{
svn_string_t *value = apr_palloc(pool, sizeof(*value));
apr_size_t vallen;
void *valbuf;
/* Get the length of the value */
SVN_ERR(svn_cstring_atoi(&parsed_len, buf + 2));
vallen = parsed_len;
/* Again, 1 extra byte for the null termination. */
valbuf = apr_palloc(pool, vallen + 1);
SVN_ERR(svn_io_file_read_full2(srcfile,
valbuf, vallen,
&num_read, NULL, pool));
((char *) valbuf)[vallen] = '\0';
/* Suck up extra newline after val data */
SVN_ERR(svn_io_file_getc(&c, srcfile, pool));
if (c != '\n')
return svn_error_create(SVN_ERR_MALFORMED_FILE, NULL, NULL);
value->data = valbuf;
value->len = vallen;
/* The Grand Moment: add a new hash entry! */
apr_hash_set(hash, keybuf, keylen, value);
}
else
{
return svn_error_create(SVN_ERR_MALFORMED_FILE, NULL, NULL);
}
}
else
{
return svn_error_create(SVN_ERR_MALFORMED_FILE, NULL, NULL);
}
} /* while (1) */
}
/* Return the next *HUNK from a PATCH in APR_FILE.
* If no hunk can be found, set *HUNK to NULL.
* Set IS_PROPERTY to TRUE if we have a property hunk. If the returned HUNK
* is the first belonging to a certain property, then PROP_NAME and
* PROP_OPERATION will be set too. If we have a text hunk, PROP_NAME will be
* NULL. If IGNORE_WHITESPACE is TRUE, lines without leading spaces will be
* treated as context lines. Allocate results in RESULT_POOL.
* Use SCRATCH_POOL for all other allocations. */
static svn_error_t *
parse_next_hunk(svn_diff_hunk_t **hunk,
svn_boolean_t *is_property,
const char **prop_name,
svn_diff_operation_kind_t *prop_operation,
svn_patch_t *patch,
apr_file_t *apr_file,
svn_boolean_t ignore_whitespace,
apr_pool_t *result_pool,
apr_pool_t *scratch_pool)
{
static const char * const minus = "--- ";
static const char * const text_atat = "@@";
static const char * const prop_atat = "##";
svn_stringbuf_t *line;
svn_boolean_t eof, in_hunk, hunk_seen;
apr_off_t pos, last_line;
apr_off_t start, end;
apr_off_t original_end;
apr_off_t modified_end;
svn_linenum_t original_lines;
svn_linenum_t modified_lines;
svn_linenum_t leading_context;
svn_linenum_t trailing_context;
svn_boolean_t changed_line_seen;
enum {
noise_line,
original_line,
modified_line,
context_line
} last_line_type;
apr_pool_t *iterpool;
*prop_operation = svn_diff_op_unchanged;
/* We only set this if we have a property hunk header. */
*prop_name = NULL;
*is_property = FALSE;
if (apr_file_eof(apr_file) == APR_EOF)
{
/* No more hunks here. */
*hunk = NULL;
return SVN_NO_ERROR;
}
in_hunk = FALSE;
hunk_seen = FALSE;
leading_context = 0;
trailing_context = 0;
changed_line_seen = FALSE;
original_end = 0;
modified_end = 0;
*hunk = apr_pcalloc(result_pool, sizeof(**hunk));
/* Get current seek position -- APR has no ftell() :( */
pos = 0;
SVN_ERR(svn_io_file_seek(apr_file, APR_CUR, &pos, scratch_pool));
/* Start out assuming noise. */
last_line_type = noise_line;
iterpool = svn_pool_create(scratch_pool);
do
{
svn_pool_clear(iterpool);
/* Remember the current line's offset, and read the line. */
last_line = pos;
SVN_ERR(readline(apr_file, &line, NULL, &eof, APR_SIZE_MAX,
iterpool, iterpool));
/* Update line offset for next iteration. */
pos = 0;
SVN_ERR(svn_io_file_seek(apr_file, APR_CUR, &pos, iterpool));
/* Lines starting with a backslash are comments, such as
* "\ No newline at end of file". */
if (line->data[0] == '\\')
{
if (in_hunk &&
((!*is_property &&
strcmp(line->data, "\\ No newline at end of file") == 0) ||
(*is_property &&
strcmp(line->data, "\\ No newline at end of property") == 0)))
{
char eolbuf[2];
apr_size_t len;
apr_off_t off;
apr_off_t hunk_text_end;
/* Comment terminates the hunk text and says the hunk text
* has no trailing EOL. Snip off trailing EOL which is part
* of the patch file but not part of the hunk text. */
off = last_line - 2;
SVN_ERR(svn_io_file_seek(apr_file, APR_SET, &off, iterpool));
len = sizeof(eolbuf);
SVN_ERR(svn_io_file_read_full2(apr_file, eolbuf, len, &len,
&eof, iterpool));
if (eolbuf[0] == '\r' && eolbuf[1] == '\n')
hunk_text_end = last_line - 2;
else if (eolbuf[1] == '\n' || eolbuf[1] == '\r')
hunk_text_end = last_line - 1;
else
hunk_text_end = last_line;
if (last_line_type == original_line && original_end == 0)
original_end = hunk_text_end;
else if (last_line_type == modified_line && modified_end == 0)
modified_end = hunk_text_end;
else if (last_line_type == context_line)
{
if (original_end == 0)
original_end = hunk_text_end;
if (modified_end == 0)
modified_end = hunk_text_end;
break;
}
SVN_ERR(svn_io_file_seek(apr_file, APR_SET, &pos, iterpool));
}
continue;
}
if (in_hunk)
{
char c;
static const char add = '+';
static const char del = '-';
if (! hunk_seen)
{
/* We're reading the first line of the hunk, so the start
* of the line just read is the hunk text's byte offset. */
start = last_line;
}
c = line->data[0];
if (original_lines > 0 && modified_lines > 0 &&
((c == ' ')
/* Tolerate chopped leading spaces on empty lines. */
|| (! eof && line->len == 0)
/* Maybe tolerate chopped leading spaces on non-empty lines. */
|| (ignore_whitespace && c != del && c != add)))
{
/* It's a "context" line in the hunk. */
hunk_seen = TRUE;
original_lines--;
modified_lines--;
if (changed_line_seen)
trailing_context++;
else
leading_context++;
last_line_type = context_line;
}
else if (original_lines > 0 && c == del)
{
/* It's a "deleted" line in the hunk. */
hunk_seen = TRUE;
changed_line_seen = TRUE;
/* A hunk may have context in the middle. We only want
trailing lines of context. */
if (trailing_context > 0)
trailing_context = 0;
original_lines--;
last_line_type = original_line;
}
else if (modified_lines > 0 && c == add)
{
/* It's an "added" line in the hunk. */
hunk_seen = TRUE;
changed_line_seen = TRUE;
/* A hunk may have context in the middle. We only want
trailing lines of context. */
if (trailing_context > 0)
trailing_context = 0;
modified_lines--;
last_line_type = modified_line;
}
else
{
if (eof)
{
/* The hunk ends at EOF. */
end = pos;
}
else
{
/* The start of the current line marks the first byte
* after the hunk text. */
end = last_line;
}
if (original_end == 0)
original_end = end;
if (modified_end == 0)
modified_end = end;
break; /* Hunk was empty or has been read. */
}
}
else
{
if (starts_with(line->data, text_atat))
{
/* Looks like we have a hunk header, try to rip it apart. */
in_hunk = parse_hunk_header(line->data, *hunk, text_atat,
iterpool);
if (in_hunk)
{
original_lines = (*hunk)->original_length;
modified_lines = (*hunk)->modified_length;
*is_property = FALSE;
}
}
else if (starts_with(line->data, prop_atat))
{
/* Looks like we have a property hunk header, try to rip it
* apart. */
in_hunk = parse_hunk_header(line->data, *hunk, prop_atat,
iterpool);
if (in_hunk)
{
original_lines = (*hunk)->original_length;
modified_lines = (*hunk)->modified_length;
*is_property = TRUE;
}
}
else if (starts_with(line->data, "Added: "))
{
SVN_ERR(parse_prop_name(prop_name, line->data, "Added: ",
result_pool));
if (*prop_name)
*prop_operation = svn_diff_op_added;
}
else if (starts_with(line->data, "Deleted: "))
{
SVN_ERR(parse_prop_name(prop_name, line->data, "Deleted: ",
result_pool));
if (*prop_name)
*prop_operation = svn_diff_op_deleted;
}
else if (starts_with(line->data, "Modified: "))
{
SVN_ERR(parse_prop_name(prop_name, line->data, "Modified: ",
result_pool));
if (*prop_name)
*prop_operation = svn_diff_op_modified;
}
else if (starts_with(line->data, minus)
|| starts_with(line->data, "diff --git "))
/* This could be a header of another patch. Bail out. */
break;
}
}
/* Check for the line length since a file may not have a newline at the
* end and we depend upon the last line to be an empty one. */
while (! eof || line->len > 0);
svn_pool_destroy(iterpool);
if (! eof)
/* Rewind to the start of the line just read, so subsequent calls
* to this function or svn_diff_parse_next_patch() don't end
* up skipping the line -- it may contain a patch or hunk header. */
SVN_ERR(svn_io_file_seek(apr_file, APR_SET, &last_line, scratch_pool));
if (hunk_seen && start < end)
{
(*hunk)->patch = patch;
(*hunk)->apr_file = apr_file;
(*hunk)->leading_context = leading_context;
(*hunk)->trailing_context = trailing_context;
(*hunk)->diff_text_range.start = start;
(*hunk)->diff_text_range.current = start;
(*hunk)->diff_text_range.end = end;
(*hunk)->original_text_range.start = start;
(*hunk)->original_text_range.current = start;
(*hunk)->original_text_range.end = original_end;
(*hunk)->modified_text_range.start = start;
(*hunk)->modified_text_range.current = start;
(*hunk)->modified_text_range.end = modified_end;
}
else
/* Something went wrong, just discard the result. */
*hunk = NULL;
return SVN_NO_ERROR;
}
/* A helper for reading a line of text from a range in the patch file.
*
* Allocate *STRINGBUF in RESULT_POOL, and read into it one line from FILE.
* Reading stops either after a line-terminator was found or after MAX_LEN
* bytes have been read. The line-terminator is not stored in *STRINGBUF.
*
* The line-terminator is detected automatically and stored in *EOL
* if EOL is not NULL. If EOF is reached and FILE does not end
* with a newline character, and EOL is not NULL, *EOL is set to NULL.
*
* SCRATCH_POOL is used for temporary allocations.
*/
static svn_error_t *
readline(apr_file_t *file,
svn_stringbuf_t **stringbuf,
const char **eol,
svn_boolean_t *eof,
apr_size_t max_len,
apr_pool_t *result_pool,
apr_pool_t *scratch_pool)
{
svn_stringbuf_t *str;
const char *eol_str;
apr_size_t numbytes;
char c;
apr_size_t len;
svn_boolean_t found_eof;
str = svn_stringbuf_create_ensure(80, result_pool);
/* Read bytes into STR up to and including, but not storing,
* the next EOL sequence. */
eol_str = NULL;
numbytes = 1;
len = 0;
found_eof = FALSE;
while (!found_eof)
{
if (len < max_len)
SVN_ERR(svn_io_file_read_full2(file, &c, sizeof(c), &numbytes,
&found_eof, scratch_pool));
len++;
if (numbytes != 1 || len > max_len)
{
found_eof = TRUE;
break;
}
if (c == '\n')
{
eol_str = "\n";
}
else if (c == '\r')
{
eol_str = "\r";
if (!found_eof && len < max_len)
{
apr_off_t pos;
/* Check for "\r\n" by peeking at the next byte. */
pos = 0;
SVN_ERR(svn_io_file_seek(file, APR_CUR, &pos, scratch_pool));
SVN_ERR(svn_io_file_read_full2(file, &c, sizeof(c), &numbytes,
&found_eof, scratch_pool));
if (numbytes == 1 && c == '\n')
{
eol_str = "\r\n";
len++;
}
else
{
/* Pretend we never peeked. */
SVN_ERR(svn_io_file_seek(file, APR_SET, &pos, scratch_pool));
found_eof = FALSE;
numbytes = 1;
}
}
}
else
svn_stringbuf_appendbyte(str, c);
if (eol_str)
break;
}
if (eol)
*eol = eol_str;
if (eof)
*eof = found_eof;
*stringbuf = str;
return SVN_NO_ERROR;
}
