Archive dump filenames contain colons, breaking cross-platform compatibility and backup integrity

[andreypfau]

Summary

Current TON node releases generate archive dump filenames containing a colon (:) in shard identifiers.
Example:

packages/arch0000/archive.00000.0:8000000000000000.pack

This format is invalid or non-portable across common storage targets (Windows, S3, OCI volumes, NTFS-mounted containers). The issue breaks interoperability and violates standard filesystem constraints.

Environment

• Affected components: validator-engine, all archive dump utilities
• Version: any node version that introduced per-shard archive segmentation (starting from commit: b8999be)
• Reproducible: always

Steps to Reproduce

1. Start validator-engine
2. Observe output filenames under the archive directory.
3. Attempt to:

• sync archives to S3 or GCS;
• store under Windows NTFS, Apple File System (APFS) or CIFS mount;
• restore using standard tooling (rsync, scp, backup scripts).

Actual Result

Filenames contain colons and are rejected or silently mangled:

• : replaced with _ or %3A by tools depending on implementation;
• On Windows and S3, object write fails outright;
• On macOS : replaced with /
• Incremental backup scripts (tar, zip) skip affected files.

Expected Result

Archive naming should be POSIX- and S3-compatible, portable across environments and reversible back to shard identifiers without lossy transformations.

Impact

• Cross-platform incompatibility: Windows, macOS, S3, and container volumes cannot store files with colons.
• Non-reversible backups: Restoring archives changes shard identifiers, breaking deterministic reconstruction.
• Toolchain breakage: rsync, tar, zstd archivers, and CI pipelines fail on invalid paths.
• Ecosystem risk: third-party operators and automated tooling (indexers, explorers, validators) become environment-dependent.

Proposed Fix

Backward-compatible normalization of filenames:

1. get_package_file_name: replace : with _ when writing new archive packages
   Example: archive.00000.0_8000000000000000.pack
2. ArchiveSlice::add_package: accept both forms

• If : found -> load package as before
• If _ found -> treat identically

3. Migration script: optional one-time rename for existing archives
4. Guideline update: note the safe character set for all future produced file names.

Justification

This is not stylistic. It violates the invariants of reproducible archival and cross-platform determinism:

• POSIX and S3 are baseline targets for archive node operators.
• File naming must satisfy both. A colon in filenames fails that requirement.

References

• AWS S3 Object Key Naming Rules
• NTFS Reserved Characters
• The POSIX portable file name character set