xdr: comprehensive XDR decoding optimization

[tamirms]

Regenerate XDR bindings and update handwritten code with multiple performance optimizations for XDR decoding.

Key changes:

1. Buffer-based decoding

   • xdr.Unmarshal now takes []byte instead of io.Reader
   • Direct buffer access eliminates io.Reader dispatch overhead
   • Use xdr.Decoder directly for primitive decoding in strkey package
   • New Decoder.Decode() method simplifies decode calls

2. Union value-type optimization

   • Primitive union arms (Bool, Int32, Uint32, Int64, Uint64, etc.) stored as values instead of pointers, reducing allocations
   • Complex union arms (ScVec, ScMap, structs) remain as pointers
   • Direct field access for union arms instead of accessor methods

3. Generated accessor improvements

   • GetX() accessors return (value, bool) for optional handling
   • MustX() accessors for cases where arm is known to be set

4. Object pooling support

   • XDR types support pooling via Pool[T] generic type
   • Enables reuse of decoded objects to reduce GC pressure

Performance improvement (LedgerCloseMeta decode, 100 varied ledgers):

Without pooling:

• Latency: 5.31ms → 3.45ms (-35%)
• Throughput: 277 MB/s → 426 MB/s (+54%)
• Memory: 8.06 MB → 7.46 MB (-7.5%)
• Allocations: 107.7k → 90.7k (-16%)

With pooling:

• Latency: 5.31ms → 2.58ms (-51%)
• Throughput: 277 MB/s → 570 MB/s (+106%)
• Memory: 8.06 MB → 2.64 MB (-67%)
• Allocations: 107.7k → 33.6k (-69%)

🤖 Generated with Claude Code

[socket-security]

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	github.com/​stellar/​go-xdr@​v0.0.0-20231122183749-b53fb00bcac2 ⏵ v0.0.0-20260106211653-82e681bfd3f3	+1

View full report

[Shaptic]

Are there going to be xdrgen changes to match the new decoding patterns?