Mercurial > octave-nkf
changeset 14835:821708f96efd
strread.m: more robust word parsing and style improvements
| author | Philip Nienhuis <prnienhuis@users.sf.net> |
|---|---|
| date | Tue, 03 Jul 2012 21:58:23 +0200 |
| parents | 619fedc6ea61 |
| children | a1e1f914ae79 |
| files | scripts/io/strread.m |
| diffstat | 1 files changed, 78 insertions(+), 60 deletions(-) [+] |
line wrap: on
line diff
--- a/scripts/io/strread.m Tue Jul 03 08:50:43 2012 -0700 +++ b/scripts/io/strread.m Tue Jul 03 21:58:23 2012 +0200 @@ -226,7 +226,7 @@ switch (lower (varargin{n})) case "bufsize" ## We could synthesize this, but that just seems weird... - warning ('strread: property "bufsize" is not implemented'); + warning ("strread: property 'bufsize' is not implemented"); case "commentstyle" comment_flag = true; switch (lower (varargin{n+1})) @@ -259,7 +259,7 @@ case "emptyvalue" numeric_fill_value = varargin{n+1}; case "expchars" - warning ('strread: property "expchars" is not implemented'); + warning ("strread: property 'expchars' is not implemented"); case "whitespace" white_spaces = varargin{n+1}; if (strcmp (typeinfo (white_spaces), "sq_string")) @@ -281,10 +281,10 @@ elseif (ischar (varargin{n+1})) empty_str = varargin(n+1); else - error ('strread: "treatasempty" value must be string or cellstr'); + error ("strread: 'treatasempty' value must be string or cellstr"); endif otherwise - warning ('strread: unknown property "%s"', varargin{n}); + warning ("strread: unknown property '%s'", varargin{n}); endswitch endfor @@ -297,10 +297,10 @@ ## Determine the number of words per line as a first guess. Forms ## like %f<literal>) (w/o delimiter in between) are fixed further on format = strrep (format, "%", " %"); - fmt_words = regexp (format, '[^ ]+', 'match'); - ## Format conversion specifiers following literals w/o space/delim - ## in between are separate now. Separate those w trailing literals - idy2 = find (! cellfun ("isempty", strfind (fmt_words, "%"))); + fmt_words = regexp (format, '[^ ]+', "match"); + + ## Find position of conversion specifiers (they start with %) + idy2 = find (! cellfun ("isempty", regexp (fmt_words, '^%'))); ## Check for unsupported format specifiers errpat = '(\[.*\]|[cq]|[nfdu]8|[nfdu]16|[nfdu]32|[nfdu]64)'; @@ -308,12 +308,15 @@ error ("strread: %q, %c, %[] or bit width format specifiers are not supported yet."); endif + ## Format conversion specifiers following literals w/o space/delim + ## in between are separate now. Separate those w trailing literals a = strfind (fmt_words(idy2), "%"); - b = regexp (fmt_words(idy2), '[nfdus]', 'end'); + b = regexp (fmt_words(idy2), '[nfdus]', "end"); for jj = 1:numel (a) + ## From right to left to avoid losing track ii = numel (a) - jj + 1; if (! (length (fmt_words{idy2(ii)}) == b{ii}(1))) - ## Fix format_words + ## Split fmt_words(ii) into % conv specifier and trailing literal fmt_words(idy2(ii)+1 : end+1) = fmt_words(idy2(ii) : end); fmt_words{idy2(ii)} = fmt_words{idy2(ii)}(a{ii} : b{ii}(1)); fmt_words{idy2(ii)+1} = fmt_words{idy2(ii)+1}(b{ii}+1:end); @@ -333,7 +336,7 @@ ## Remove comments in str if (comment_flag) ## Expand 'eol_char' here, after option processing which may have set value - comment_end = regexprep (comment_end, 'eol_char', eol_char); + comment_end = regexprep (comment_end, "eol_char", eol_char); cstart = strfind (str, comment_start); cstop = strfind (str, comment_end); ## Treat end of string as additional comment stop @@ -363,7 +366,8 @@ endif if (! isempty (white_spaces)) - ## For numeric fields, whitespace is always a delimiter, but not for text fields + ## For numeric fields, whitespace is always a delimiter, but not for text + ## fields if (isempty (strfind (format, "%s"))) ## Add whitespace to delimiter set delimiter_str = unique ([white_spaces delimiter_str]); @@ -399,16 +403,15 @@ str = str(2:end); endif ## Check for single delimiter followed/preceded by whitespace - ## FIXME: Double strrep on str is enormously expensive of CPU time. - ## Can this be eliminated if (! isempty (delimiter_str)) dlmstr = setdiff (delimiter_str, " "); rxp_dlmwsp = sprintf ("( [%s]|[%s] )", dlmstr, dlmstr); str = regexprep (str, rxp_dlmwsp, delimiter_str(1)); endif + ## Wipe leading and trailing whitespace on each line (it may be + ## delimiter too) ## FIXME: Double strrep on str is enormously expensive of CPU time. ## Can this be eliminated - ## Wipe leading and trailing whitespace on each line (it may be delimiter too) if (! isempty (eol_char)) str = strrep (str, [eol_char " "], eol_char); str = strrep (str, [" " eol_char], eol_char); @@ -418,13 +421,12 @@ ## Split 'str' into words words = split_by (str, delimiter_str, mult_dlms_s1, eol_char); if (! isempty (white_spaces)) - ## Trim leading and trailing white_spaces - ## FIXME: Is this correct? strtrim clears what matches isspace(), not - ## necessarily what is in white_spaces. + ## Trim leading and trailing 'white_spaces'. All whitespace has + ## been converted to space above words = strtrim (words); endif num_words = numel (words); - ## First guess at number of lines in file (ignoring leading/trailing literals) + ## First guess at nr. of lines in file (ignoring leading/trailing literals) num_lines = ceil (num_words / num_words_per_line); ## Replace TreatAsEmpty char sequences by empty strings @@ -436,9 +438,8 @@ endif ## fmt_words has been split properly now, but words{} has only been split on - ## delimiter positions. - ## As numeric fields can also be separated by whitespace, more splits may be - ## needed. + ## delimiter positions. As numeric fields can also be separated by + ## whitespace, more splits may be needed. ## We also don't know the number of lines (as EndOfLine may have been set to ## "" (empty) by the caller). ## @@ -447,6 +448,8 @@ ## B: Leading literals (<literal>%f) w/o delimiter in between. ## C. Skipping leftover parts of specified skip fields (%*N ) ## Some words columns may have to be split further to fix these. + ## To find out, we'll match fmt_words to the words array to see what + ## needs to be done. fwptr tracks which {fmt_words}# starts in what {words}# ## Find indices and pointers to possible literals in fmt_words idf = cellfun ("isempty", strfind (fmt_words, "%")); @@ -454,7 +457,7 @@ idg = ! cellfun ("isempty", regexp (fmt_words, '%\*?\d')); idy = find (idf | idg); ## Find indices to numeric conversion specifiers - idn = ! cellfun ("isempty", regexp (fmt_words, "%[dnfu]")); + idn = ! cellfun ("isempty", regexp (fmt_words, '%[dnfu]')); ## If needed, split up columns in three steps: if (! isempty (idy)) @@ -464,12 +467,16 @@ ## 1. Assess "period" in the split-up words array ( < num_words_per_line). ## Could be done using EndOfLine but that prohibits EndOfLine = "" option. ## Alternative below goes by simply parsing a first grab of words - ## and counting words until the fmt_words array is exhausted: - iwrd = 1; iwrdp = 0; iwrdl = length (words{iwrd}); + ## and matching fmt_words to words until the fmt_words array is exhausted. + ## iwrd: ptr to current analyzed word; iwrdp: ptr to pos before analyzed char + iwrd = 1; iwrdp = 0; iwrdl = length (words{1}); + fwptr = zeros (1, numel (fmt_words)); ii = 1; while ii <= numel (fmt_words) nxt_wrd = 0; + ## Keep track of which words nr. every fmt_words{} is (starts) in. + fwptr(ii) = iwrd; if (idf(ii)) ## Literal expected @@ -493,8 +500,9 @@ elseif (idg(ii)) ## Fixed width specifier (%N or %*N): read just a part of word - iwrdp += floor ... - (str2double (fmt_words{ii}(regexp(fmt_words{ii}, '\d') : end-1))); + sw = regexp (fmt_words{ii}, '\d', "once"); + ew = regexp (fmt_words{ii}, '[nfuds]') - 1; + iwrdp += floor (str2double (fmt_words{ii}(sw:ew))); if (iwrdp > iwrdl) ## Match error. Field extends beyond word boundary. warning ... @@ -532,10 +540,11 @@ if (iwrd > numel (words)) ## Apparently EOF; assume incomplete row already at L.1 of data ii = numel (fmt_words); - elseif (ii < numel (fmt_words)) + elseif (ii < numel (fmt_words) && iwrd <= numel (words)) iwrdl = length (words{iwrd}); endif endif + ++ii; endwhile @@ -544,12 +553,11 @@ num_lines = ceil (num_words / words_period); ## 2. Pad words array so that it can be reshaped - tmp_lines = ceil (num_words / words_period); - num_words_padded = tmp_lines * words_period - num_words; + num_words_padded = num_lines * words_period - num_words; if (num_words_padded) words = [words'; cell(num_words_padded, 1)]; endif - words = reshape (words, words_period, tmp_lines); + words = reshape (words, words_period, num_lines); ## 3. Do the column splitting on rectangular words array icol = 1; ii = 1; # icol = current column, ii = current fmt_word @@ -582,6 +590,7 @@ (@(x) substr(x, e(1)+1, length(x)-e(1)), words(icol, jptr), ... "UniformOutput", false); words(icol, jptr) = fmt_words{ii}; + fwptr = [fwptr(1:ii) (++fwptr(ii+1:end))]; else if (! idg(ii) && ! isempty (strfind (fmt_words{ii-1}, "%s"))) @@ -589,11 +598,19 @@ warning ("Ambiguous '%s' specifier next to literal in column %d", icol); elseif (idg(ii)) ## Current field = fixed width. Strip into icol, rest in icol+1 - wdth = floor (str2double (fmt_words{ii}(regexp(fmt_words{ii}, ... - '\d') : end-1))); + sw = regexp (fmt_words{ii}, '\d', "once"); + ew = regexp (fmt_words{ii}, '[nfuds]') - 1; + wdth = floor (str2double (fmt_words{ii}(sw:ew))); words(icol+1, jptr) = cellfun (@(x) x(wdth+1:end), words(icol,jptr), "UniformOutput", false); - words(icol, jptr) = strtrunc (words(icol, jptr), wdth); + if (isempty ([words(icol+1, :){:}])) + ## Apparently split wasn't needed as turns out to cover + ## entire column. So delete column again + words(icol+1, :) = []; + else + words(icol, jptr) = strtrunc (words(icol, jptr), wdth); + fwptr = [fwptr(1:ii) (++fwptr(ii+1:end))]; + endif else ## FIXME: this assumes char(254)/char(255) won't occur in input! clear wrds; @@ -610,6 +627,7 @@ char(254), fmt_words{ii}), char(255)); ## Former trailing literal may now be leading for next specifier --ii; + fwptr = [fwptr(1:ii) (++fwptr(ii+1:end))]; endif endif endif @@ -617,9 +635,7 @@ else ## Conv. specifier. Peek if next fmt_word needs split from current column if (ii < num_words_per_line) - if (idf(ii+1) && (! isempty (strfind (words{icol, 1}, fmt_words{ii+1})))) - --icol; - elseif (idg(ii+1)) + if (fwptr(ii) == fwptr(ii+1)) --icol; endif endif @@ -678,10 +694,12 @@ varargout{k} = data.'; k++; case {"%0", "%1", "%2", "%3", "%4", "%5", "%6", "%7", "%8", "%9"} - nfmt = strsplit (fmt_words{m}(2:end-1), '.'); + sw = regexp (fmt_words{m}, '\d', "once"); + ew = regexp (fmt_words{m}, '[nfudsq]') - 1; + nfmt = strsplit (fmt_words{m}(2:ew), "."); swidth = str2double (nfmt{1}); - switch fmt_words{m}(end) - case {"d", "u", "f", "n%"} + switch fmt_words{m}(ew+1) + case {"d", "u", "f", "n"} n = cellfun ("isempty", data); ### FIXME - erroneously formatted data lead to NaN, not an error ### => ReturnOnError can't be implemented for numeric data @@ -774,7 +792,7 @@ %! a = rand (10, 1); %! b = char (randi ([65, 85], 10, 1)); %! for k = 1:10 -%! str = sprintf ('%s %.6f %s\n', str, a(k), b(k)); +%! str = sprintf ("%s %.6f %s\n", str, a(k), b(k)); %! endfor %! [aa, bb] = strread (str, "%f %s"); %! assert (a, aa, 1e-6); @@ -785,19 +803,19 @@ %! a = rand (10, 1); %! b = char (randi ([65, 85], 10, 1)); %! for k = 1:10 -%! str = sprintf ('%s %.6f %s\n', str, a(k), b(k)); +%! str = sprintf ("%s %.6f %s\n", str, a(k), b(k)); %! endfor %! aa = strread (str, "%f %*s"); %! assert (a, aa, 1e-6); %!test -%! str = sprintf ('/* this is\nacomment*/ 1 2 3'); +%! str = sprintf ("/* this is\nacomment*/ 1 2 3"); %! a = strread (str, "%f", "commentstyle", "c"); %! assert (a, [1; 2; 3]); %!test %! str = "# comment\n# comment\n1 2 3"; -%! [a, b] = strread (str, '%n %s', 'commentstyle', 'shell', 'endofline', "\n"); +%! [a, b] = strread (str, "%n %s", "commentstyle", "shell", "endofline", "\n"); %! assert (a, [1; 3]); %! assert (b, {"2"}); @@ -922,12 +940,12 @@ %! assert (c, [0.94; 0.87], 0.01) %!test -%! [a, b] = strread (['Empty 1' char(10)], 'Empty%s %f'); +%! [a, b] = strread (["Empty 1" char(10)], "Empty%s %f"); %! assert (a{1}, '1'); %! assert (b, NaN); %!test -%! [a, b] = strread (['Empty' char(10)], 'Empty%f %f'); +%! [a, b] = strread (["Empty" char(10)], "Empty%f %f"); %! assert (a, NaN); %! assert (b, NaN); @@ -940,17 +958,17 @@ %% Unsupported format specifiers %!test -%!error <format specifiers are not supported> strread ('a', '%c') -%!error <format specifiers are not supported> strread ('a', '%*c %d') -%!error <format specifiers are not supported> strread ('a', '%q') -%!error <format specifiers are not supported> strread ('a', '%*q %d') -%!error <format specifiers are not supported> strread ('a', '%[a]') -%!error <format specifiers are not supported> strread ('a', '%*[a] %d') -%!error <format specifiers are not supported> strread ('a', '%[^a]') -%!error <format specifiers are not supported> strread ('a', '%*[â] %d') -%!error <format specifiers are not supported> strread ('a', '%d8') -%!error <format specifiers are not supported> strread ('a', '%*d8 %s') -%!error <format specifiers are not supported> strread ('a', '%f64') -%!error <format specifiers are not supported> strread ('a', '%*f64 %s') -%!error <format specifiers are not supported> strread ('a', '%u32') -%!error <format specifiers are not supported> strread ('a', '%*u32 %d') +%!error <format specifiers are not supported> strread ("a", "%c") +%!error <format specifiers are not supported> strread ("a", "%*c %d") +%!error <format specifiers are not supported> strread ("a", "%q") +%!error <format specifiers are not supported> strread ("a", "%*q %d") +%!error <format specifiers are not supported> strread ("a", "%[a]") +%!error <format specifiers are not supported> strread ("a", "%*[a] %d") +%!error <format specifiers are not supported> strread ("a", "%[^a]") +%!error <format specifiers are not supported> strread ("a", "%*[^a] %d") +%!error <format specifiers are not supported> strread ("a", "%d8") +%!error <format specifiers are not supported> strread ("a", "%*d8 %s") +%!error <format specifiers are not supported> strread ("a", "%f64") +%!error <format specifiers are not supported> strread ("a", "%*f64 %s") +%!error <format specifiers are not supported> strread ("a", "%u32") +%!error <format specifiers are not supported> strread ("a", "%*u32 %d")