Skip to content

Latest commit

 

History

History
206 lines (125 loc) · 10.2 KB

CHANGES.md

File metadata and controls

206 lines (125 loc) · 10.2 KB

Changes

Next

3.3.0

3.2.0

  • NokogiriReader has a "nokogiri.strict_mode" setting. Set to true or string 'true' to ask Nokogori to parse in strict mode, so it will immediately raise on ill-formed XML, instead of nokogiri's default to do what it can with it. traject#226

  • SolrJsonWriter

    • Utility method delete_all! sends a delete all query to the Solr URL endpoint. traject#227

    • Allow basic auth configuration of the default http client via solr_writer.basic_auth_user and solr_writer.basic_auth_password. traject#231

3.1.0

Added

  • Context#add_output is added, convenient for custom ruby code.

      each_record do |record, context|
     context.add_output "key", something_from(record)
  end

    traject#220

  • SolrJsonWriter

    • Class-level indexer configuration, for custom indexer subclasses, now available with class-level configure method. Warning, Indexers are still expensive to instantiate though. traject#213

    • SolrJsonWriter has new settings to control commit semantics. solr_writer.solr_update_args and solr_writer.commit_solr_update_args, both have hash values that are Solr update handler query params. traject#215

    • SolrJsonWriter has a delete(solr-unique-key) method. Does not currently use any batching or threading. traject#214

    • SolrJsonWriter, when MaxSkippedRecordsExceeded is raised, it will have a #cause that is the last error, which resulted in MaxSkippedRecordsExceeded. Some error reporting systems, including Rails, will automatically log #cause, so that's helpful. traject#216

    • SolrJsonWriter now respects a solr_writer.http_timeout setting, in seconds, to be passed to HTTPClient instance. traject#219

    • Only runs thread pool shutdown code (and logging) if there is a solr_writer.batch_size greater than 0. Keep it out of the logs if it was a no-op anyway.

    • Logs at DEBUG level every time it sends an update request to solr

  • Nokogiri dependency for the NokogiriReader increased to ~> 1.9. When using Jruby each_record_xpath, resulting yielded documents may have xmlns declarations on different nodes than in MRI (and previous versions of nokogiri), but we could find now way around this with nokogiri >= 1.9.0. The documents should still be semantically equivalent for namespace use. This was necessary to keep JRuby Nokogiri XML working with recent Nokogiri releases. traject#209

  • LineWriter guesses better about when to auto-close, and provides an optional explicit setting in case it guesses wrong. (thanks @justinlittman) traject#211

  • Traject::Indexer will now use a Logger(-compatible) instance passed in in setting 'logger' traject#217

3.0.0

Changed/Backwards Incompatibilities

  • JRuby traject no longer includes traject-marc4j_reader as a dependency or default reader, although it may provide faster MARC-XML reading on JRuby. To use it manually, see https://github.com/traject/traject-marc4j_reader . See traject#187

  • map_record now returns nil if record was skipped.

  • The Traject::Indexer class no longer includes marc-specific settings and modules.

    • If you are using command-line traject, this should make no difference to you, as command-line now defaults to the new Traject::Indexer::MarcIndexer with those removed things.
    • If you are using Traject::Indexer programmatically and want those features, switch to using Traject::Indexer::MarcIndexer.
    • If neccessary, as a hopefully temporary backwards compat shim, call Traject::Indexer.legacy_marc_mode!, which injects the old marc-specific behavior into Traject::Indexer again, globally and permanently.

  • Traject::Indexer::Settings no longer has it's own global defaults, Instead it can be given a set of defaults with #with_defaults, usually right after instantiation. To support different defaults for different Indexers.

  • SolrJsonWriter now assumes an /update/json convenience url is available in solr instead of trying to verify it. If you are using an older Solr (before 4?) or otherwise want a different update url, just use setting solr.update_url

Added

  • Traject::Indexer#configure is available, and recommended instead of raw instance_eval. It just does an instance_eval, but is clearer and safer for future changes.

  • traject command line can now take multiple input files. And underlying it, Traject::Indexer#process can take an array of input streams.

  • There is now a built-in mode for XML source records, see docs at xml.md

  • new setting mapping_rescue is available, to supply custom logic for handling errors. See docs at settings.md

  • Call Traject::ThreadPool.disable_concurrency! to force all pool sizes to be 0, and work to be performed inline. All threading will be disabled.

  • to_field can now take an array as a first argument, to send values to multiple fields mentioned, eg:

    to_field ["field1", "field2"], extract_marc("240")

  • to_field can take multiple transformation procs (all with the same form). traject#153

  • There is a new set of standard transformation macros included in Traject::Indexer, from Traject::Macros::Transformation. It includes an extraction of previous/existing arguments from marc_extract, along with some additional stuff. , in [Traject::Macros::Transformations]. traject#154

    • This is the new preferred way to do post-processing with the marc_extract options, but the existing options are not deprecated and there is no current plan for them to be removed.

    • before:

      to_field "some_field", extract_marc("800",
                        translation_map: "marc_800_map",
                        allow_duplicates: true,
                        first: true,
                        default: "default value")

    • now preferred:

      to_field "some_field", extract_marc("800", allow_duplicates: true),
    translation_map("marc_800_map"),
    first_only,
    default("default value")

      (still need allow_duplicates: true cause extract_marc defaults to false, but see also unique macro)

    • So, these transformation steps can now be used with non-MARC formats as well. See also new transformation macros: strip, split, append, prepend, gsub, and transform. And for MARC use, trim_punctuation.

  • Traject::Indexer new api, for more convenient programmatic/embedded use.

    • Traject::Indexer.new takes a block for config

    • Traject::Indexer#process_record

    • Traject::Indexer#process_with

    • Traject::Indexer#complete and #run_after_processing_steps public API.

  • Traject::SolrJsonWriter#flush, flush to solr without closing, may be useful for direct programmatic use.

  • Traject::Indexer sub-classes can implement a #source_record_id_proc, which is passed to Context, for source-format-specific logic for getting an ID to use in logging.

  • command line takes an -i flag for choice of indexer.

2.3.4

  • Totally internal change to provide easier hooks into indexing process

2.3.3

  • Further squash use of capture-variabels ('$1', etc.) to try to work around the non-thread-safety of regexp in ruby
  • Fix a bug in trim_punctuation where trailing periods were being eliminated even if there was a short string before them (e.g., 'Jr.')
  • Begin to reorganize tests, starting with the Marc21 macros

2.3.2

  • Change to extract_marc to work around a threadsafe problem in JRuby/MRI where regexps were unsafely shared between threads. (@codeforkjeff)
  • Make trim-punctuation safe for non-just-ASCII text (thanks to @dunn and @redlibrarian)

2.3.1

  • Update README with more info about new nil-related options

2.3.0

  • Allow nil values, empty fields, and deduplication

    This adds three new settings (all of whose defaults reflect current behavior)

    • allow_nil_values (default: false). Allow nil values to be sent on to the writer
    • allow_duplicate_values (default: true). Allow duplicate values. Set to false to force only unique values.
    • allow_empty_fields (default: false). Default behavior is that the output hash doesn't even contain keys for a to_field which doesn't produce any values. Set to true to pass empty fields on to the writer (with the value being an empty array)

2.2.1

  • Had inadvertently broken use of arrays as extract_marc specifications. Fixed.

2.2.0

  • Change DebugWriter to be more forgiving (and informative) about missing record-id fields
  • Automatically require DebugWriter for easier use on the command line
  • Refactor MarcExtractor to be easier to read
  • Fix .travis file to actually work, and target more recent rubies.

2.1.0

  • update some docs (typos)
  • Make the indexer's writer r/w so it can be set at runtime (#110)
  • Allow extract_marc to be callable from anywhere (#111)
  • Add doc instructions/examples for programmatic Indexer use
  • Much better error reporting; easier to find which record went wrong

2.0.2

  • Guard against assumption of MARC data when indexing using SolrJsonWriter (#94)
  • For MARC Records, try to use the production date when available (#93)

2.0.1

  • Fix bad constant in logging (#91)

2.0.0

  • Compatible with MRI/RBX
  • Default to SolrJsonWriter
  • Release separate MRI/JRuby gems

1.0

  • First release