GSoC Third Week Progress

Picking Up Momentum After Convocation Week

The second week of the coding period was comparatively slower on my side as I was occupied with the convocation process and related activities at my university. However, once that wrapped up, work picked up pace again during the third week with a stronger focus on implementation, documentation updates, and API consistency improvements across data.table.

A large part of the work this week involved understanding how users interact with :=, repeated execution patterns, and formula interfaces like dcast().

Documenting := Side Effects in Repeated Execution (Issue #7409, PR #7786)

One of the main contributions this week was improving documentation around side effects caused by := during repeated execution.

The discussion originated from subtle behavior involving tests that execute multiple optimization levels, where in-place modification through := can unexpectedly affect subsequent runs unless copy() is used carefully.

My work focused on documenting these behaviors across:

• datatable-reference-semantics.Rmd
• assign.Rd
• test.Rd

The updates clarify:

• How reference semantics interact with repeated execution
• Why copy() becomes important in certain testing workflows
• How optimization-based repeated evaluation can expose side effects

The goal was to make these behaviors easier to understand for contributors writing tests and users working with by-reference modification patterns.

Supporting Raw Expressions in dcast(subset=) (Issue #4325, PR #7789)

Another contribution this week involved improving consistency in the dcast() interface.

Previously, the subset= argument primarily relied on .() syntax. I updated the logic so that dcast() can now also accept raw logical expressions directly inside subset=.

For example, users can now write more natural expressions similar to base R semantics without requiring additional wrapping syntax.

This improves:

• Consistency with base::subset()
• Overall usability of dcast()
• Readability of user code

The implementation was intentionally kept minimal to preserve backward compatibility while extending the existing interface behavior.

Proposed Fixes and Ongoing Discussions

Alongside merged and open PR work, I also spent time analyzing additional issues and discussing potential implementation approaches with maintainers.

Some ongoing discussions include:

• Investigating missing realloc warnings during repeated := operations inside functions (Issue #1633)
• Exploring support for a keep argument in tstrsplit() (Issue #5250)

For both issues, I proposed potential approaches and will begin implementation work once the design direction is finalized with maintainers.

This week involved a strong mix of consistency improvements, documentation refinement, and design discussions around user-facing behavior. Looking forward to continuing work on both usability and internal consistency across the data.table ecosystem.