Developer's job board deduplication bug wrongly flagged 166 of 174 listings as duplicates

·4 views

A developer building GigsPaid, a board for paid coding gigs, found that its near-duplicate filter was discarding most incoming job posts, leaving the site with only 18 listings. Replaying a single scan showed the filter kept just 8 of 174 tech gigs. The root cause was a flawed simhash implementation that produced only 31 usable bits and had several bits that barely varied across documents. As a result, unrelated job posts averaged an 11.1-bit Hamming distance, below the duplicate threshold of 12, so most pairs were treated as duplicates. After replacing the hash with 64 bits from two independent hash functions and using the standard threshold of 3, the same scan stored all 174 listings and production listings rose above 100.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Offline-first Flutter design for robot operators and field technicians

The article addresses how robots and edge devices in warehouses, factories, and outdoor settings need apps that keep working when connectivity drops. It recommends a local cache and command outbox behind a connectivity-aware repository, classifying actions as read-only, safely retryable, or non-replayable. Retryable operations should use idempotency keys, while commands that cannot be safely queued require explicit user confirmation. On reconnection, the app should refresh authoritative device state before enabling actions, and motion commands must never be replayed automatically. The guidance also covers performance measurement at p50 and p95/p99 latency, bounded queues, and testing on actual target hardware.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Profiling Flutter Apps for Smart Glasses and Robotics Latency and Memory

This tutorial positions Flutter as the operator-facing layer, with wearable SDKs, Jetson services, ROS 2 nodes, and NVIDIA-accelerated inference handling device-specific and compute-heavy work. Its stated goals are predictable latency, bounded memory, useful observability, and graceful degradation rather than maximum message throughput. It warns that safety-critical control must stay on the robot, since a Flutter app is not a hard real-time controller. Recommended practice includes profiling on physical hardware in profile mode, tracking p50, p95, and p99 latency, and using bounded queues, latest-value caches, and explicit overflow policies. It also advises keeping parsing out of build methods and benchmarking isolates before using them for CPU-heavy work.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Six GnuTLS CI test jobs can fail without breaking the pipeline

A static analysis of GnuTLS's GitLab CI configuration found that six of its fifteen test-running jobs are set to allow failure, so a failing test still shows the pipeline as passed. The affected jobs include Windows (mingw) builds under Wine, a ThreadSanitizer suite, and several cross-architecture runs under QEMU. The project's other nine test jobs do hard-fail the pipeline, which suggests the exceptions were not intended to be permanent. Two of the jobs carry a comment about a single step failing in a rootless container, but allow_failure applies to the whole job, so real regressions could go unnoticed. The four cross-architecture jobs have no stated reason at all, and the big-endian ppc64le and s390x jobs are the only ones testing those architectures. The article proposes forgiving only infrastructure errors while still failing on genuine assertion failures.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Better Support Documentation Helps SaaS Teams Resolve Customer Issues Faster

Support tickets often fail to preserve the findings and troubleshooting steps needed for future cases, which forces teams to start investigations from scratch. The article uses a hypothetical SaaS app where customers' PDF uploads fail with an HTTP 413 (Content Too Large) error, illustrating why detailed ticket information such as file type, size, browser, and error code matters. It explains that HTTP 413 signals a request exceeds what the server will process, but the limit may be enforced at the application, web server, proxy, or another infrastructure layer. The piece argues that well-documented tickets help support teams respond consistently and give engineers the context to investigate and reproduce technical problems. It also presents a troubleshooting workflow and emphasizes turning resolutions into reusable knowledge.

0 comments Read more at DEV Community