Abstract: Embodiments improve software defect diagnosis. Analytic focus is automatically walked back from an initial symptomatic diagnostic context to a previous diagnostic context that is closer to underlying causes. Diagnosis may obtain diagnostic artifacts such as traces or dumps, extract diagnostic context, decompile executables, lookup likely causes based on symptoms, scan logs, and submit diagnostic context to software analysis services. An analysis service may perform static analysis, security testing, symptom-pair lookups, or antipattern scanning, for example, and may include a neural network or other machine learning model, for example. Root causes are culled from analysis results and identified to a software developer. Changes to mitigate the defect's impact are suggested in some cases. Thus, the software developer receives debugging leads without manually navigating through all the tool interfaces or unrelated details of diagnostic contexts.

Abstract: Methods, systems, and computer program products for using a confidence measure to automatically identify a diagnostic analyzer that applies to a diagnostic artifact. A plurality of diagnostic analyzers are each configured to analyze diagnostic artifacts relating to prior executions of software entities. A confidence measure is calculated for each diagnostic analyzer. Each confidence measure indicates a likelihood that the diagnostic analyzer applies to a particular diagnostic artifact. Calculating each confidence measure comprises applying one or more heuristics specific to the diagnostic analyzer against the particular diagnostic artifact, with an outcome of application of each heuristic contributing to the confidence measure for the respective diagnostic analyzer.

Abstract: Methods, systems, and computer program products for using a confidence measure to automatically identify a diagnostic analyzer that applies to a diagnostic artifact. A plurality of diagnostic analyzers are each configured to analyze diagnostic artifacts relating to prior executions of software entities. A confidence measure is calculated for each diagnostic analyzer. Each confidence measure indicates a likelihood that the diagnostic analyzer applies to a particular diagnostic artifact. Calculating each confidence measure comprises applying one or more heuristics specific to the diagnostic analyzer against the particular diagnostic artifact, with an outcome of application of each heuristic contributing to the confidence measure for the respective diagnostic analyzer.

Abstract: Embodiments improve software defect diagnosis. Analytic focus is automatically walked back from an initial symptomatic diagnostic context to a previous diagnostic context that is closer to underlying causes. Diagnosis may obtain diagnostic artifacts such as traces or dumps, extract diagnostic context, decompile executables, lookup likely causes based on symptoms, scan logs, and submit diagnostic context to software analysis services. An analysis service may perform static analysis, security testing, symptom-pair lookups, or antipattern scanning, for example, and may include a neural network or other machine learning model, for example. Root causes are culled from analysis results and identified to a software developer. Changes to mitigate the defect's impact are suggested in some cases. Thus, the software developer receives debugging leads without manually navigating through all the tool interfaces or unrelated details of diagnostic contexts.

Abstract: Embodiments provide improved diagnosis of software defects. Static analysis services and other source-based diagnostic tools and techniques are applied even when the source code underlying software is unavailable. Diagnosis obtains diagnostic artifacts, extracts diagnostic context from the artifacts, decompiles to get source, and submits decompiled source to a source-based software analysis service. The analysis service may be a static analysis tool, an antipattern scanner, or a machine learning model trained on source code, for example. The diagnostic context may also guide the analysis, e.g., by localizing decompilation or prioritizing possible causes. Likely causes are culled from analysis results and identified to a software developer. Changes to mitigate the defect's impact are suggested. Thus, the software developer receives debugging leads without providing source code for the defective program, and without manually navigating through a decompiler and through the analysis services.