compaction

type Config struct {
	// TargetFileSizeBytes is the desired output file size.
	// Default: table.WriteTargetFileSizeBytesDefault (512 MB).
	TargetFileSizeBytes int64

	// MinFileSizeBytes is the lower bound for a file to be considered "optimal".
	// Files smaller than this are candidates for compaction.
	// Default: 75% of TargetFileSizeBytes.
	MinFileSizeBytes int64

	// MaxFileSizeBytes is the upper bound. Files larger than this are never rewritten
	// (unless they exceed DeleteFileThreshold).
	// Default: 180% of TargetFileSizeBytes.
	MaxFileSizeBytes int64

	// MinInputFiles is the minimum number of files in a group to justify rewriting.
	// Groups with fewer files are dropped from the plan.
	// Default: DefaultMinInputFiles.
	MinInputFiles uint

	// DeleteFileThreshold is the minimum number of delete files associated with
	// a data file to force it into compaction regardless of file size.
	// Default: 5.
	DeleteFileThreshold int

	// PackingLookback controls how many open bins the packer considers
	// before evicting. Higher values produce better packing at the cost
	// of memory. Default: DefaultPackingLookback.
	PackingLookback uint

	// PreserveDeadEqualityDeletes, when true, retains equality delete
	// files that are provably dead after the rewrite. The cleanup
	// predicate matches the v2 reader (see scanner.go
	// matchEqualityDeletesToData and DecideDeadEqualityDeletes): an
	// eq-delete is dead iff no surviving applicable data file has
	// seq < eq-delete.seq, where "applicable" means same partition
	// tuple OR either side has empty partition. SpecID is NOT part of
	// the predicate.
	//
	// Zero value (false) is the recommended default: dead eq-deletes are
	// expunged during the rewrite commit, which keeps manifest fanout
	// bounded under sustained CDC workloads where eq-deletes accumulate
	// one-per-snapshot. Set to true only if a downstream consumer
	// depends on the historical eq-delete files surviving in the live
	// snapshot's manifests (rare).
	//
	// Honored only in atomic mode (RewriteDataFilesOptions.PartialProgress=false).
	// The CLI / library caller is responsible for translating this flag
	// into the snapshot walk + RewriteDataFilesOptions.ExtraDeleteFilesToRemove
	// — see CollectDeadEqualityDeletes.
	PreserveDeadEqualityDeletes bool
}

type Group struct {
	// PartitionKey is an opaque grouping key derived from partition values.
	// Empty string for unpartitioned tables.
	PartitionKey string

	// Tasks are the input FileScanTasks that should be compacted together.
	Tasks []table.FileScanTask

	// TotalSizeBytes is the sum of data file sizes in this group.
	TotalSizeBytes int64

	// DeleteFileCount is the total number of delete files across all tasks.
	DeleteFileCount int
}

type Plan struct {
	// Groups are the sets of files to compact, each within a single partition.
	Groups []Group

	// SkippedFiles is the count of files that are already optimal or oversized.
	SkippedFiles int

	// TotalInputFiles is the total number of files scanned.
	TotalInputFiles int

	// TotalInputBytes is the total size of all scanned files.
	TotalInputBytes int64

	// EstOutputFiles is the estimated number of output files after compaction.
	EstOutputFiles int

	// EstOutputBytes is the estimated total size of output files.
	// This is an upper-bound estimate: actual output may be smaller because
	// deleted rows are removed and Parquet compression/encoding is more
	// effective on larger files with better column statistics.
	EstOutputBytes int64
}

type SurvivorSurvey struct {
	EmptyPartMinSeq int64
	PartMinSeq      map[string]int64
}