Python 3.15 Changes File Defaults and Import Timing: What to Test
Python 3.15.0 became a stable release on October 9, bringing a change that can reach existing applications without a source-code edit: UTF-8 is now the default text encoding. The release also introduces explicit lazy imports, which let developers postpone loading a module until it is needed. For teams planning an upgrade, those features call for two separate checks: how files are decoded and when initialization happens.
The distinction matters. The encoding default changes with the interpreter; lazy imports require opting in. A useful migration therefore starts by testing existing behavior, then evaluates deferred imports as a separate performance change.
Check the files your tests rarely see
Python's upgrade notes say that text I/O without an explicit encoding, such as open('report.txt'), now uses UTF-8 independently of the system environment. An explicit encoding still takes precedence. That makes behavior more consistent across machines, but does not convert old files into UTF-8.
PEP 686, the design behind the change, identifies Windows as the main compatibility concern because many Unix environments already use UTF-8. It warns that applications depending on a different default can encounter decoding errors, garbled text, or silent corruption.
Consider a hypothetical customer-import job that receives a file from older business software. An ASCII-only test fixture may look fine while a real customer's accented name exposes the encoding mismatch. Add representative files from the actual producing system to the migration tests, including non-ASCII characters, and compare the decoded values rather than merely checking that the job exits successfully.
Choose an encoding according to the file's documented format. For UTF-8 data, specify encoding='utf-8'; when the operating system's locale encoding is deliberately required, Python supports encoding='locale'. PEP 686 recommends finding implicit-encoding dependencies and then testing with UTF-8 mode enabled.
Lazy imports move work—and failures—to first use
The new lazy keyword can prefix a module-level import, for example lazy import json. Loading is deferred until the imported name is accessed. The documentation also supports lazy from imports and explains that the keyword cannot be used inside functions, classes, or try/except/finally blocks. Existing ordinary imports remain eager unless an opt-in mechanism changes them.
This can help a command-line application whose help screen otherwise loads dependencies for every subcommand. It also changes the point at which a missing or broken dependency becomes visible: an import failure can emerge when the selected feature first runs. PEP 810 describes both that delayed error timing and the deferral of module-level side effects.
That second consequence deserves a targeted test. Suppose importing a plugin also registers its handlers. Making the import lazy can leave the registry empty until something accesses the imported name. PEP 810 recommends explicit initialization or discovery for such patterns. Keep necessary startup work eager unless the application has another deliberate way to trigger it.
Make the upgrade measurable
Start with the existing application's tests under Python 3.15, without changing its import policy. For encoding investigation, Python provides -X warn_default_encoding; -X utf8=0 disables UTF-8 mode when you need a comparison. Use those options with the project's normal test runner, then repeat under the new default. The command-line reference documents both switches.
Next, measure imports before selecting a candidate for deferred loading. Python's -X importtime reports time spent in individual imports and their dependencies. For a CLI, compare a quick path such as help with a real subcommand that uses the proposed lazy dependency. Record total completion time as well as startup time: moving a wait to the first operation may be worthwhile, but it is still a wait.
Finally, exercise the failure path and any plugin discovery after the change. A successful help command is weak evidence that an export, upload, or report-generation command still works. Keep the interpreter upgrade and import optimization in separate reviewable changes so a regression has a smaller search area.
Python 3.15 also brings a faster experimental JIT compiler, new built-in types, and profiling tools. Its release notes describe performance gains for specific benchmark configurations; they do not establish a speedup for your application. The more useful first result is a tested upgrade with correct file handling, followed by a measured improvement to a startup path that users actually notice.
Photo: Christina Morillo / Pexels. Used under the Pexels License.
Sources
// keep reading