dsv 0.12.9 → 0.12.10

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 7171a2d2de4dba26f3e304ca3e0c216e9bb093a36cef1420629e9f936e644889
-  data.tar.gz: 4ccf92f6000c7e5b9d82315babc17c816edd3df7587583eb79f2c16eb19125cb
+  metadata.gz: f88116d5530f25575975779af46933ddd7c67af651b2c61a3894b237b000950a
+  data.tar.gz: cd03ed4c4b47c12044387544346d8be13ac3f256f1263f135302134bf1ce429d
 SHA512:
-  metadata.gz: 4727efe3e9571299914d4b6731169ac5a70eabd475176f06cb084dba35cfcc038420faf4740ad01bbedb9caea1b13ba21ed40a058f2699ca6f4fedf9ba47a0b1
-  data.tar.gz: 6b6143f85e9c6106344fed384228433057eff0641a38eff9e6f823a2d905cca416284b7068f80ff26529feb02045fe09f323e689614012bd267d0783aec469ba
+  metadata.gz: 0a6e2477a6ff620bba7e5317af9ec1482f0effa853f3ac6f6074e1460170f201b0649db16c1ba67b87621546b986948678d4f6610f2623e29766cb9d948c227d
+  data.tar.gz: b9ad0b92bcf1ec1caee1604217005250411ac318014ceafb9384ceb70ae276f5d2e457e097a287dcbb5c9d9b668144e8a2c35194d9e7e2a9d28377c5aa3bf8b9

data/CHANGELOG

@@ -1,5 +1,15 @@
 # dsv/CHANGELOG
 
+## 20261103
+
+0.12.10: ~ TODO, now grouped by release; + ROADMAP.md.
+
+1. ~ TODO: the jobs decided, by release: the fixes left for 0.12.x; modes as declarations in 0.13.0; streaming in 0.14.0; what 1.0 needs; speed and new options in 1.x.
+2. ~ TODO: 11, strict:, and 15, the malformed-quoting option, superseded by field_count: and :rfc_4180; 13, chunks, by a streaming each; 1, 6, 7 and the undecided parts of 2, to ROADMAP.md.
+3. + ROADMAP.md: the design, the boundary for 1.0, the version rule, the release sequence, and the themes after 1.0 with what is undecided in each.
+4. ~ lib/DSV/VERSION.rb: /0.12.9/0.12.10/.
+
+
 ## 20261028
 
 0.12.9: + DSV::UnterminatedQuote, raised where a quoted field does not close.

data/TODO

@@ -1,25 +1,46 @@
 # TODO
 
-## 0.12.x, before the gem is published
+## 0.12.x, fixes
 
-1. A path is told from content by the call, not by a guess: .read, .foreach, .open and the other path-taking class methods take a path, .parse takes text, new keeps the guess with a switch to say which, and a path under a read mode that names no file raises rather than being read as text.
-2. Selection as a keyword on the calls, only: or select:, columns: being taken; the positional form kept or retired; selected_columns: renamed to match.
-3. The instance-level header_row accessor renamed headers?, the class-level header_row now being the row.
-4. << to append a row and generate to build a string, the two writing calls a CSV user reaches for first.
+1. The class methods close what they open, an early exit included, and without a block return an Enumerator which opens the source afresh on each pass, as CSV.foreach does; collect, select, reject and detect without a block no longer raise.
+2. as_array: with a column selection: each('name') no longer raises TypeError, and read('name', as_array: true) gives only the columns selected. as_array: in the README, as the way to get Arrays as CSV gives them.
+3. Reading honours quote: given as a String, as writing does; DSV.parse_line honours quote:.
+4. A space as the column separator splits on each space, not as String#split(' ') does.
+5. :double back to 0.12.4's reading, undoing 0.12.5 item 2: a row not starting with a quote split plainly, one starting with a quote split on quote-separator-quote, and one which leaves a lone quote in a field read in full as wholly quoted. The README back to 0.12.3's wording.
+6. :single read, as :double is with a single quote.
+7. filter, a class-level alias of select.
+8. A String under a write mode taken as a path, so that a file not yet there can be written by name.
+9. The internals private.
 
-## 0.13.0, speed
+## 0.13.0, modes as declarations
 
-5. The saved-up speed change, with the evidence in the reconciliation: two string predicates in place of four regex matches per piece in split_csv; rows built by zip.to_h; a benchmark script in the repository against stdlib CSV, and the README's speed sentence written from its output. The scanner-based parser on the branch scanner-parser is judged here.
+10. An unknown quote mode, file mode or option raises ArgumentError, where each is ignored today.
+11. :mixed, the name of the default.
+12. A mode raises on a contradiction which costs nothing to see, every row alike, header included: under :double, a row not starting with a quote, and a bare field in a row read in full. Declaring :mixed reads such a row.
 
-## Features, after
+## 0.14.0, streaming
 
-6. Guessing, as a mode that runs once at construction on a sample of the first rows and feeds the same chooser as an explicit option, scanner only: the column separator, the one whose count is the same on every sampled row; whether the first row is a header, its columns' types differing from the rows beneath in the same way across columns; and, where there is no header, a header made from the data, each column named for what its values look like, integer, decimal, date, email, URL, phone, with a positional fallback, column_0, for text and a suffix where two columns share a class. Names come from what values look like, not what they mean; anything more is a schema, and another library's.
-7. Templated ingestion: a template: option taking a whole-row Regexp with a capture per field, the group names being the column names, one match per row; an Array of separators as its writable special case, compiled into such a Regexp at construction and invertible for writing where a general template is not.
-8. The duplicate-header modes as an option: keep all as an Array, the default; index, a, a.1; first; last; error. All derivations of the same columns map at header time, none costing anything per row.
-9. Writing options: minimal quoting, a field quoted only where it holds a separator, a quote or a row terminator; a row terminator inside a field kept, quoted, or folded to a space; the encoding declared and a transcode offered with any loss marked.
-10. Symbols and Strings interchangeable for column names, in a selection, a columns: list and a row's keys; today a String is required.
-11. A strict: option, off by default, that raises on a row with more or fewer fields than the header; today a short row gets only the columns it has and a long row's excess lands under nil.
-12. A line number in each parse error, found only when the error is raised, from the source's position at that moment, as UnterminatedQuote finds its; never counted as it goes, since counting costs time on every row for the sake of the rare one.
-13. Chunked processing of a large file: a bounded number of rows at a time handed to a block, so that nothing needs the whole file in memory and a partial result can be acted on before the rest is read.
-14. Checks, all of them after the above: strict: (13), column types, and whatever else proves wanted; correct and published comes first.
-15. The rule for malformed quoting as an option, its name to be settled, perhaps one with strict: (11): the default, as now; a stricter one, under which a quote that does not open a field raises too, and :double raises on a row which is not what it promises rather than reading it as unspecified quoting does; and a lenient one, never raising, text after a closing quote kept and a field which does not close running to the end of the source.
+13. each yields each row as it is read and returns its receiver, so that detect and break stop reading and memory does not grow with the file; rows is no longer filled by each. Chunks come from each_slice. A source read once, start to end, with no rewind, so that a pipe or $stdin can be read: the header from the first row read, and the line of an error found by seeking only where the source can; a rewind left to r+ rewriting and first_row called alone.
+
+## 1.0
+
+14. A corpus in the tests, a day of NEMWEB's dispatch files trimmed and the RFC 4180 cases, each checked against stdlib CSV's rows.
+15. A benchmark script against stdlib CSV, and the README's speed sentence written from its output.
+16. README: every mode by example, and a table matching each to stdlib CSV's and Python's settings.
+
+## 1.x, speed, results unchanged
+
+17. A per-row fast path for the default: split, the fields bare or wrapped in two quotes unwrapped, any other row read in full.
+18. as_array: returning the split fields, rather than building a Hash and then an Array from it.
+19. The splitter chosen once, at construction, as row_renderer is for writing.
+20. Two string predicates in place of four regex matches per piece; rows built by zip.to_h; the scanner-based parser, at the tag scanner-parser-20261006, judged against the rest.
+
+## 1.x, features
+
+21. :rfc_4180, a quote mode reading what RFC 4180 allows as :mixed does and raising on what it forbids.
+22. field_count:, :header or an Integer, off by default: a row of another width raises.
+23. Writing: minimal quoting, a field quoted only where it holds a separator, a quote or a row terminator; a row terminator inside a field kept, quoted, or folded to a space; the encoding declared and a transcode offered with any loss marked.
+24. A line number in each parse error, found only when the error is raised, as UnterminatedQuote finds its, where the source can seek. line_numbers:, off by default, counts lines as it reads instead, for any source a pipe included: the read loop chosen once, at construction, as row_renderer is for writing, so that without it no row pays for the count.
+25. The instance-level header_row renamed headers?, kept as an alias; << to append a row and generate to build a string.
+26. Symbols and Strings interchangeable for column names, in a selection, a columns: list and a row's keys.
+27. The duplicate-header modes as an option: keep all as an Array, the default; index, a, a.1; first; last; error.

data/lib/DSV/VERSION.rb

@@ -2,5 +2,5 @@
 # DSV::VERSION
 
 class DSV
-  VERSION = '0.12.9'
+  VERSION = '0.12.10'
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: dsv
 version: !ruby/object:Gem::Version
-  version: 0.12.9
+  version: 0.12.10
 platform: ruby
 authors:
 - thoran