[mstackhouse]

Closes #78. Targets yyjson-native (#98), not main — it builds directly on the native NDJSON reader and writer added there.

What DSJC is

The specification is deliberately minimal: a DSJC file is the Dataset NDJSON content of a dataset as a zLib stream, with no header, signature or wrapper of its own. .dsjc extension, MIME type application/vnd.cdisc.dataset-json.compressed, level 9 and a 15-bit window recommended.

So it sits directly on the NDJSON work already on #98.

Three functions

write_dataset_dsjc(x, file, level = 9)	Returns a raw vector when no file is given, since the format is binary
read_dataset_dsjc(file)	Path, URL, or raw vector
validate_dataset_dsjc(x)	Decompresses and validates the content as Dataset NDJSON

Implementation note worth a reviewer's attention

memCompress(type = "gzip") does emit a conformant zLib stream despite the name — but at a fixed level 6, and the specification recommends 9. So this uses zlib directly (PKG_LIBS = -lz; R requires zlib on every platform it builds on). That also allows rows to be deflated as they are serialized, so writing a large dataset never holds its uncompressed NDJSON in memory — which is what the specification's memory-efficiency section asks for.

If the extra link is unwelcome, memCompress would work at level 6 and remove ~120 lines of C, at the cost of buffering the whole payload.

Verified

• Header is 0x78 0xDA — zLib, level 9
• Decompressed content is byte-identical to what write_dataset_ndjson() writes to a file
• Interoperability both directions: streams written here decompress with R's own zlib at every level 0–9, and a stream compressed elsewhere reads back correctly
• Corrupt and truncated input report rather than crash (invalid distances set, compressed stream ended before it was complete)
• Numbers survive the compressed path bit-exactly, using the same no-tolerance identical() check as the other formats
• URL support is binary-safe — read_from_url() reads lines, which would corrupt a stream, so there is a separate read_raw_from_url(). Tested over file://, which exercises the same path as http(s) without needing a network

26MB / 300k rows
level 1: write 0.21s  read 0.06s  11.9 MB (45.8%)
level 6: write 0.86s  read 0.06s  11.0 MB (42.1%)
level 9: write 1.05s  read 0.06s  10.9 MB (41.9%)

Level 9 costs 5x the write time for 4% smaller output. Kept as the default because the specification recommends it, with the trade-off documented.

Also here

inst/extdata now carries all three representations of the same dataset — dm.json, dm.ndjson, dm.dsjc — generated from dm.json in data-raw/data.R so they cannot drift. A test asserts they read back to the same dataset, including that dm.dsjc really is dm.ndjson compressed. The read_dataset_ndjson() and validate_dataset_ndjson() examples now read a real file rather than writing one first.

The Getting Started vignette gains a "Compressed files" section.

32 tests. R CMD check --as-cran: 0 errors, 0 warnings, 0 notes.

---

• Place Closes #<insert_issue_number> into the beginning of your Pull Request Title
• Code is formatted according to the tidyverse style guide
• Updated relevant unit tests or have written new unit tests
• If you removed/replaced any function and/or function parameters, did you fully follow the deprecation guidance? — no removals here
• Update to all relevant roxygen headers and examples
• Run devtools::document()
• Address any updates needed for vignettes and/or templates
• Update NEWS.md — deferred to the release cleanup pass
• Build pkgdown site and check new functions appear on the Reference page
• Address or fix all lintr warnings and errors
• Run R CMD check locally and address all errors and warnings
• Link the issue in the Development Section on the right hand side.
• Address all merge conflicts and resolve appropriately
• Pat yourself on the back for a job well done!

[codecov]

Codecov Report

❌ Patch coverage is 93.66197% with 9 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (yyjson-native@a44cb5c). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
src/write.c	90.56%	5 Missing ⚠️
R/dsjc.R	92.68%	3 Missing ⚠️
R/utils.R	87.50%	1 Missing ⚠️

Additional details and impacted files

@@               Coverage Diff                @@
##             yyjson-native      #99   +/-   ##
================================================
  Coverage                 ?   97.76%           
================================================
  Files                    ?       14           
  Lines                    ?     1031           
  Branches                 ?        0           
================================================
  Hits                     ?     1008           
  Misses                   ?       23           
  Partials                 ?        0           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.