mangle

Author: Ludovic HENRY ludovic.henry@qti.qualcomm.com
Date: 2026-06-17
Scope: RISC-V (riscv64/linux) support status for mangle
Audience: Technical leadership, resource allocation strategy
Verification policy: Every claim is cross-referenced to a primary upstream source. Items that could not be verified against a second source are marked [NEEDS VERIFICATION].

1. Project Overview

Mangle is a Datalog-based deductive database programming language implemented as a Go library. It provides a logic programming model layered over a factstore, intended for rule-driven inference over structured data. The canonical source is at codeberg.org/TauCeti/mangle-go; github.com/google/mangle is a mirror. A separate Rust implementation lives at codeberg.org/TauCeti/mangle-rs.

The project originated inside Google and was released as open source, but the README explicitly states: “Mangle started out as an open source project released by Google but is now developed independently. It is not and never was an officially supported Google product.” Google has no ongoing ownership or support obligation.

Governance: Informal, single-maintainer. CONTRIBUTING.md names Burak Emir as the lead; all submissions require PR review except those by Burak. There is no foundation membership (not Apache, CNCF, Linux Foundation, or RISE Project). License: Apache-2.0.

Corporate sponsors: None identified. Burak Emir uses a personal Proton email address and shows no current corporate affiliation. The second-most active contributor, Alexandre Philibert, also uses a personal email. The project is effectively a solo open-source effort.

Community culture on new ports: No port tiering policy, no PLATFORMS.md, no SUPPORT.md. The project delegates architecture portability entirely to the Go toolchain. There is no precedent for community discussion of architecture-specific work because no such work has ever been required.

Repository stats: 2986 stars, 154 forks, last pushed 2026-08-05. [NEEDS VERIFICATION] - star/fork counts from a single source (search findings).

2. Port History and Upstreaming Timeline

Date	Event	Source
-	No RISC-V commits, issues, PRs, or code references exist in google/mangle	GitHub Issues API, Commits API, Code Search: 0 results for “riscv”, “riscv64”, “risc-v”
-	No architecture-specific code exists for any platform (amd64, arm64, or riscv64)	Full repo tree inspection: no .S files, no SIMD dispatch, no JIT, no GOARCH build constraints
-	riscv64 support is implicit via Go toolchain (`GOARCH=riscv64`) since the project’s initial release	Go has supported linux/riscv64 as a target since Go 1.14

There is no port history to report. The project is pure Go with no architecture-specific code, so “porting” is not a concept that applies. GOOS=linux GOARCH=riscv64 go build ./... has worked since Go added riscv64 support, with no action required by the mangle project.

Key contributors (by commit count, all platforms): Burak Emir (~61 of ~95 commits), Alexandre Philibert (~18 commits), Christian G. Warden (3), Daniel Ostrow (2), several one-commit contributors. No contributor is affiliated with a semiconductor or RISC-V ecosystem company.

3. Upstream Support Tier

Mangle has no formal tier policy. The Go toolchain’s tier classification for linux/riscv64 applies transitively.

Criterion	amd64	arm64	riscv64
Builds	Yes	Yes	Yes (via `GOARCH=riscv64`)
CI	None	None	None
Official binaries	None	None	None
Release-blocking tests	None	None	None
Formal tier	N/A	N/A	N/A

The project has no CI for any architecture. There are no release binaries for any architecture. All four releases (v0.1.0 through v0.4.0) have zero GitHub release assets. This is not a riscv64-specific gap; it is the baseline state of the project for all platforms.

4. Technical Architecture and RISC-V-Specific Subsystems

Mangle is a pure Go implementation of a Datalog interpreter with a fact store backend, ANTLR4-based parser, and an interactive REPL. There is no JIT compiler, no SIMD acceleration, no cryptographic implementation, no garbage collector override, and no hand-written assembly in any component.

Code search results for riscv, riscv64, aarch64, x86_64, GOARCH, simd, assembly, vfloat32m1_t, rvv against google/mangle all return 0 results. The full directory tree contains only pure Go packages: analysis/, ast/, builtin/, engine/, factstore/, functional/, interpreter/, parse/, and others. There is a rust/ subdirectory containing only a README.md pointing to the separate mangle-rs repository.

Component	amd64	arm64	riscv64
Core interpreter (engine/)	scalar Go	scalar Go	scalar Go
Parser (antlr4-go)	scalar Go	scalar Go	scalar Go
Fact store (factstore/)	scalar Go	scalar Go	scalar Go
JIT backend	missing	missing	missing
SIMD/vectorized ops	missing	missing	missing
Assembly routines	missing	missing	missing
Crypto primitives	missing	missing	missing
GC barriers	managed by Go runtime	managed by Go runtime	managed by Go runtime

There is no riscv64-specific gap relative to amd64 or arm64. All platforms receive the same scalar Go implementation. No ISA extensions (RVV, Zba, Zbb, or otherwise) are used or relevant.

5. Build System, Cross-Compilation, and Toolchain

The build system is the standard Go toolchain exclusively. No CMake, no Make, no Bazel, no Dockerfile is present in the repository.

Required Go version: 1.25.0 (from go.mod: go 1.25.0).

Standard build:

go get -t ./...
go build ./...
go test ./...

Install interpreter binary:

GOBIN=~/bin go install codeberg.org/TauCeti/mangle-go/interpreter/mg@latest

Cross-compile for riscv64:

GOOS=linux GOARCH=riscv64 go build ./...
GOOS=linux GOARCH=riscv64 go test -c ./...

No -DUSE_X=OFF flags exist (no CMake). No known architecture-specific build flags. No CGo in the main module; cross-compilation requires no C toolchain.

QEMU is not required for building. QEMU would be needed only to execute riscv64 test binaries on a non-riscv64 host. This is not documented anywhere in the project.

Known build failures on riscv64: None identified. No open issues or PRs reference build failures on any architecture.

The Rust implementation (mangle-rs) targets wasm32-unknown-unknown primarily. Cross-compiling the Rust code to riscv64gc-unknown-linux-gnu is expected to work via cargo build --target riscv64gc-unknown-linux-gnu but this is not documented. The wasmtime dependency used in mangle-vm supports riscv64 as a host [NEEDS VERIFICATION - Codeberg anti-scraper protection prevented CI workflow inspection].

6. Feature Coverage and Gap Analysis vs arm64 and amd64

Feature	amd64	arm64	riscv64	Gap
Datalog interpreter	Full	Full	Full	None
ANTLR4 parser	Full	Full	Full	None
Fact store (in-memory and columnar)	Full	Full	Full	None
Interactive REPL (mg binary)	Full	Full	Full	None
Protocol Buffers serialization	Full	Full	Full	None
Compression (via klauspost/compress)	SIMD-accelerated	SIMD-accelerated	Scalar Go fallback	Performance gap only (see below)
JIT compilation	None	None	None	No gap (feature absent everywhere)
Floating-point correctness issues	None reported	None reported	None reported	None
NaN / IEEE 754 anomalies	None reported	None reported	None reported	None

Performance gap: github.com/klauspost/compress uses hand-written assembly for hot paths (huff0 decoder, S2 encoder/decoder, zstd FSE decoder, xxhash) on amd64 and arm64. On riscv64, it falls back to pure Go via decode_other.go / encode_go.go guarded by build tags. Throughput for zstd/S2 is expected to be 2-5x lower on riscv64 than on amd64/arm64. This matters only if mangle is deployed with heavy on-disk compression of fact stores. There is no correctness impact.

Security hardening gaps: Data not available - no security hardening analysis was performed for any platform.

7. CI/CD Infrastructure

The google/mangle repository has no CI configuration of any kind. GET /repos/google/mangle/contents/.github/workflows returns HTTP 404. There is no .gitlab-ci.yml, no Jenkinsfile, no .cirrus.yml.

The only automated tooling present is CodeQL (security scanning) and Dependabot (dependency updates), both GitHub-managed without explicit workflow YAML.

CI dimension	amd64	arm64	riscv64
Build CI	None	None	None
Test CI	None	None	None
RISE runner	No	No	No
Hardware runners	None	None	None
QEMU emulation	None	None	None

RISE Project involvement: Data not available from riseproject.dev/blog (JavaScript-rendered). The RISE wheel builder (78-package list, fully enumerated) does not include mangle. The riseproject-dev GitHub organization (30 repos scanned) contains no mangle reference.

8. Distribution and Release Status

GitHub releases: v0.1.0 through v0.4.0, all with zero release assets. Source-only releases.

PyPI: A package named mangle (v1.0.0) exists on PyPI with an x86_64-only manylinux wheel and a source sdist. This is an unrelated project. The google/mangle Go library is not distributed via PyPI.

Debian: HTTP 404 at tracker.debian.org/pkg/mangle. Not packaged.

Ubuntu Noble: No package named mangle in Ubuntu 24.04. Search returns unrelated results (demangling libraries, pkgbinarymangler).

Arch Linux RISC-V: No result at archriscv.felixc.at.

To obtain a working riscv64 binary: A user must install Go 1.25.0+, then run:

GOOS=linux GOARCH=riscv64 GOBIN=/target/bin go install codeberg.org/TauCeti/mangle-go/interpreter/mg@latest

or cross-compile from source. No pre-built binary is available through any distribution channel for any architecture.

9. Dependencies

All direct dependencies are pure Go modules with no C extensions. No dependency has a JIT, SIMD implementation, or numerics subsystem that is critical to mangle’s correctness.

Dependency	Role	riscv64 Build	riscv64 Test	riscv64 Release	Blocking Issues
Go toolchain (>=1.25.0)	Runtime, compiler, stdlib	Tier 2; linux/riscv64 builder exists	Partial - builder reliability issues (#79067, #79068, #79069 open in golang/go)	Since Go 1.14	Open builder reliability bugs in golang/go (#79067-79069); no correctness bug for mangle code
antlr4-go/antlr v4.13.1	ANTLR4 parser runtime	Yes - pure Go	Yes - no arch-specific tests	Yes	None found
klauspost/compress v1.18.6	Compression (zstd, S2, zlib)	Yes - generic Go fallback via `decode_other.go` / `encode_go.go`	No riscv64 CI (matrix: amd64/386 only)	Yes	Performance gap only: amd64/arm64 get assembly hot paths; riscv64 uses pure Go
chzyer/readline v1.5.1	Interactive REPL line editing	Yes - pure Go + golang.org/x/sys	No riscv64-specific issues found	Yes	None
golang/glog v1.2.5	Structured logging	Yes - pure Go	Yes	Yes	None
google/go-cmp v0.7.0	Deep equality (test dep)	Yes - pure Go	Yes	Yes	None
uber-go/multierr v1.11.0	Error aggregation	Yes - pure Go	Yes	Yes	None
google.golang.org/protobuf v1.36.11	Protocol Buffers (Go runtime only; no C extension)	Yes - pure Go	No riscv64 issues found	Yes	None. The C++ protobuf library is a separate project not used here.
bitbucket.org/creachadair/stringset	Immutable string sets	Yes - pure Go	Yes	Yes	None
golang.org/x/exp (indirect)	Experimental Go APIs (slices, maps)	Yes - pure Go	Yes	Yes	None
golang.org/x/sys (indirect)	OS syscall layer (used by chzyer/readline)	Yes - `asm_linux_riscv64.s`, `syscall_linux_riscv64.go` present	Yes	Yes	None

Deep dive - Go toolchain: The linux-riscv64-rva22u64-mengzhuo builder has reliability issues tracked in golang/go as #79067, #79068, and #79069 (open). A GORISCV64=g profile proposal (#74540) would add RVV support but is not yet merged. These are infrastructure issues in the Go project, not in mangle. No build correctness bug exists for mangle code on riscv64.

Deep dive - klauspost/compress: The assembly paths cover huff0 decoding, S2 encoding/decoding, zstd FSE decoding, and xxhash on amd64 and arm64. riscv64 falls back to pure Go throughout. RVV intrinsics are absent. This is a performance-only gap with no correctness impact.

No dependency has a JIT backend requiring riscv64 porting work. No dependency has unresolved riscv64 correctness bugs.

11. Known Bugs and Active Issues

All open issues in google/mangle were scanned. Zero reference RISC-V, architecture-specific behavior, floating-point correctness, or performance:

ID	Title	Status	Severity	Notes
#65	Unification persistent data structure	Open	Feature request	No arch relevance
#40	Add version number to simplecolumn file format	Open	Enhancement	No arch relevance
#36	FR: Java Wrapper?	Open	Feature request	No arch relevance
#25	Analysis should report error when multiple declarations	Open	Bug	No arch relevance
#3	Grammar railroad diagram	Open	Documentation	No arch relevance

No correctness bugs. No performance bugs. No riscv64-related issues in any state (open or closed).

12. Objections and Upstream Blockers

Technical blockers: None. The project is pure Go, and Go supports riscv64. No porting work is required.

Organizational blockers: None. The maintainer (Burak Emir) has made no statements about RISC-V and there is no policy excluding any architecture.

Stated objections: None found. No issue or PR discussion mentions RISC-V in any context.

CI addition acceptance probability: High [NEEDS VERIFICATION]. The project has no CI for any platform, so a PR adding riscv64 CI would need to establish the entire CI infrastructure from scratch. The maintainer’s low activity on infrastructure (no CI in 4 releases) suggests this may not be a priority. No governance barrier exists, but bandwidth is limited to one primary maintainer.

13. Investment Analysis

Google is a RISE Project Premier Member. No RISE-funded work on mangle has been identified in any source. RISE does not list mangle in its wheel builder or any public work item.

13.1 Functional Enablement

No functional enablement work is required. mangle compiles and runs on riscv64 with GOARCH=riscv64 go build ./... today. There are no missing features, no broken subsystems, and no architecture-specific code paths to implement.

13.2 Performance Optimization

The only performance gap is in klauspost/compress (used by mangle’s fact store for optional compression), which lacks RVV-accelerated paths for riscv64. This is upstream work in the klauspost/compress project, not in mangle itself. The impact is limited to deployments using compressed fact stores.

13.3 CI/CD Infrastructure

The project has no CI for any platform. Adding riscv64 CI means building the entire CI infrastructure from scratch. The absence is uniform across all architectures; riscv64 is not disadvantaged relative to amd64 or arm64.

13.4 Ecosystem Enablement

Not applicable. Mangle is a standalone Go library and interpreter with no dependent package ecosystem requiring separate riscv64 enablement.

13.5 Summary Table

Area	Work Item	Effort (person-weeks)	Owner	Priority
CI/CD	Add GitHub Actions CI for build and test (all platforms including riscv64)	0.5	Upstream (Burak Emir) or contributor	Low
Performance	RVV-accelerated paths in klauspost/compress for zstd/S2 hot paths	4-8	klauspost/compress upstream	Low
Distribution	Add riscv64 pre-built binary to GitHub releases	0.5	Upstream (Burak Emir) or contributor	Low

All items are Low priority. The project is functionally complete on riscv64 with no investment. The CI gap is a quality-of-life issue, not a correctness or availability blocker. The compression performance gap affects only a subset of deployments and the magnitude (2-5x on compression throughput) is irrelevant unless mangle is deployed as a high-throughput fact store server, which is not its stated use case.

Recommendation: No investment required for RISC-V functional enablement of mangle. If Qualcomm or a RISE member has a specific deployment scenario requiring riscv64 CI assurance, the CI addition is a half-week task with high acceptance probability.

14. Updates

No updates yet – initial report dated 2026-06-17.