No, using only the block API is not suitable; your cross-process long-stream case should prefer the frame API.

• The README distinguishes the block format from the frame format and says arbitrarily long files or streams use multiple blocks.
• Those blocks are organized into a frame, which provides the organization required for streaming data.
• It explicitly says interoperable LZ4 implementations must respect the frame format, directly matching a cross-process and cross-machine interoperability requirement.
• The API documentation maps lz4.h to the block API and lz4frame.h to the frame API.

The block API remains suitable when an existing protocol already manages block boundaries. The README does not say whether your transport supplies lengths, checksums, truncation detection, or dictionary identifiers, so it cannot establish that raw blocks are safe.

It depends: embedding the BSD 2-Clause library is a different licensing case from shipping the GPL-2.0-or-later CLI, so they must not be treated as one license object.

• The README explicitly says the LZ4 library uses the BSD 2-Clause license.
• It separately states that the lz4 command-line utility uses the GPL-2.0-or-later license.
• Project metadata labels the license as “Other,” so metadata alone is insufficient and the README’s library/CLI distinction matters.
• The Packaging status section says most distributions package both liblz4 and the lz4 CLI, but distribution still requires treating them separately.

Embedding only the library is therefore a different compliance path from shipping the CLI. The README does not explain static versus dynamic linking, modified CLI distribution, combined products, or source-code obligations, so it cannot provide a final legal determination.

Yes, because LZ4 is specifically designed for high-throughput compression and extremely fast decompression.

• The README states compression above 500 MB/s per core; its single-thread benchmark reaches 770 MB/s in default mode.
• The same benchmark reports 4,950 MB/s decompression, and the README says multi-core performance commonly reaches RAM-speed limits.
• The acceleration factor can trade compression ratio for higher write-side speed dynamically.
• The project provides a C library plus block and frame APIs for memory blocks and continuous streams.

Actual end-to-end throughput may still be limited by memory bandwidth, copying, or downstream processing. The README does not specify your cache data distribution or concurrency model.

Yes, provided that backup writing is not a real-time bottleneck; LZ4_HC concentrates the extra CPU cost on compression.

• The README states that LZ4_HC trades CPU time for improved compression ratio while all versions retain the same decompression speed.
• The benchmark reports about 40 MB/s compression and a 2.717 factor for LZ4 HC -9, versus 2.112 for default LZ4.
• HC -9 decompresses at 4,800 MB/s, close to the default mode’s 4,950 MB/s.
• The high-compression API is exposed through lz4hc.h, allowing direct C/C++ integration.

It is not suitable for an online path with very tight write latency. The README does not specify the backup data’s real compressibility, disk throughput, or resource curve for other HC levels.

Yes, the README explicitly lists these build paths, although the target distribution and ABI still need verification.

• The core implementation is C; the README provides a Make installation flow and says the Makefile supports parallel builds, staged installs, and standard variable conventions.
• It explicitly provides CMake, Meson, and Visual Studio build systems, covering common Linux and Windows project environments.
• The vcpkg section gives ./vcpkg install lz4, which fits teams already using vcpkg for dependency management.
• Project data identifies v1.11.0 as the latest release and reports five releases, so version and ABI compatibility still require separate checking.

The README does not specify generated library names, exported targets, compiler options, or ABI guarantees for each build system, nor whether the vcpkg port is synchronized with v1.11.0.

Yes, especially when short messages and small files have stable structure and both endpoints can synchronize the dictionary.

• The README explicitly supports dictionary compression at both API and CLI levels.
• The dictionary may be any input file, but only its final 64 KB are used, so making the file arbitrarily larger is not useful.
• It says this capability can be combined with the Zstandard Dictionary Builder to drastically improve compression for small files.
• The API entry point is lib/lz4frame.h, while the CLI uses the -D operation modifier.

This is not a universal improvement for random data or messages with unrelated structures. The README does not state whether your Python binding exposes the dictionary API, nor the dictionary distribution cost, versioning, or mismatch behavior.