qvd2parquet v0.4.0

Added

• --out-dir converts a whole folder: pass files or directories, and each
  .qvd becomes a .parquet of the same name. A failing file does not stop
  the run; every input is attempted, failures are listed at the end, and the
  exit code reports the most actionable one. --recursive descends into
  subdirectories. Two inputs that would produce the same output file are
  refused before anything is written, since --force would otherwise silently
  overwrite the first result.

• --file-workers converts several files at once and divides the decode
  workers between them, so the total stays near one per CPU. This is why folder
  conversion is built in rather than left to a shell loop: separate processes
  would each start NumCPU workers and oversubscribe the machine.

• --log writes JSON Lines, one record per file plus a summary, so a finished
  run can be queried with DuckDB or jq rather than read. Each record carries
  row and column counts, output size, elapsed time, throughput, and the quality
  gate's verdict. In batch mode --schema-report and --quality-report write
  one document per input, named after it.

• --inspect reads the XML header and the symbol tables, prints the schema a
  conversion would produce, and exits without touching the record area. The
  cost is independent of row count: on a 29 MiB, 5-million-row file it reads
  7.9 KiB and finishes in 0.01s, against 1.58s for the full conversion. All
  type policy flags apply, so the report shows what a conversion would write.
  A file the type policy rejects prints the reason plus the raw symbol profiles
  that explain it and exits 3, which makes it a cheap pre-flight check. The
  report goes to stdout; --schema-report also works in inspect mode.

• Qlik's semantic field tags are now read. A field declaring no
  NumberFormat/Type but tagged $date, $timestamp, $time or $interval
  is resolved to that type. This works for plain numeric fields carrying no
  dual display strings, which no amount of text inspection could identify, and
  it is what Qlik Sense writes. A declared type still wins over a tag.

• An empty string symbol is written as null, which is how Qlik treats it.
  The substitution is counted and reported. Pass --empty-as-null=false to
  keep "" distinct from null.

• A NaN or infinite value in a date, timestamp, time, integer or decimal
  column is written as null rather than failing the conversion. Such a
  value is not something those types can hold, and nothing is lost by nulling
  it. The substitution is counted and reported on stderr and in
  --schema-report, so it is never silent. A finite value that simply does
  not fit is still an error, because nulling it would discard real data.
  float64 columns keep NaN and infinity, which they can represent.

• A date or timestamp column is now validated when the schema is resolved,
  whatever decided its type -- the declared header, a Qlik tag, or display-string
  inference. A value that cannot be converted fails as a schema policy error
  naming the column and the value, so --inspect predicts it and no output file
  is started, instead of the conversion failing part-way through.

• --infer-dates (on by default) is the fallback for files that carry no tags:
  a column with no declared type is read as a date or timestamp when every
  display string renders its Excel-style serial value as one. The check is
  format-agnostic and accepts a neighbouring day, since the string was rendered
  in whatever timezone wrote the file and a whole-hour offset can move the
  calendar date. Serials outside roughly 1900 to 2200 are never read as dates,
  blank display strings are not evidence, and a column mixing dates with
  anything else is left alone.

  Only words that belong to a rendered date -- month and weekday names in
  English and German, meridiem markers, ordinal suffixes and timezone
  abbreviations -- may appear alongside the digits, and a month or weekday name
  that contradicts the value is rejected, so "Mon, 20 Nov 2010" beside a
  Saturday keeps its text. The ISO 8601 T between date and time is treated as
  punctuation. A clock time uses a narrower list still, since a month or weekday
  name is not something a time32 value can encode. So "Due 11/20/2010" and "20 Jan 2010"
  beside a November value both keep their text column. The list errs short: a
  word wrongly rejected costs a redundant column, whereas one wrongly accepted
  drops text that carried information.

Changed

• --dual now defaults to auto. A Qlik dual's display string is written
  as a ${name}__text column only when it carries something the numeric column
  does not. A localized number such as 1.234,56, or a date rendered beside a
  column that already encodes it, is redundant and dropped; a label such as
  Open beside 1 is kept, and the reason is reported. One informative string
  is enough to keep the column, so the default errs towards preserving data.
  --dual=numeric, text and columns still force a choice.

  Redundancy is judged against the value that will actually be written, not the
  raw payload, so a MONEY field carrying 1.234 and displaying 1.23 at
  scale 2 does not produce a text column.

• The banner now carries the year: (c) 2026, RALFORION d.o.o.

Install

Download the archive for your platform below, unpack it, and put qvd2parquet
on your PATH. Verify the download against SHA256SUMS:

shasum -a 256 -c SHA256SUMS --ignore-missing

Binaries are pure Go and statically linked, so they need no runtime
dependencies: Linux (amd64, arm64), Windows (amd64, arm64) and macOS (amd64,
arm64).