USING PYCFORGE
Ting provides six additional reviews after a conversion attempt. Every finding identifies a bounded check, its evidence and its limits. A successful conversion alone is not proof of correct behavior.
After converting, select the Ting control in the status bar. The overview summarizes the review. Select an agent tab for its findings, or open Flow Map to see the six reviews together.
Editing source or document identities, changing the supplied bundle or mode, or starting another conversion retires the old report. Convert again for current feedback. The report records the exact request, output and reasoning-policy fingerprints.
| Agent | What its feedback concerns |
|---|---|
| Syntax | Exact source/request binding and fresh source parsing. Parsing does not establish conversion support. |
| Semantics | Revalidation of Typed analysis obligations. Complete Python/C behavioral equivalence remains unproved. |
| Ownership | Typed scope and file-cleanup model relationships. General alias safety, allocation failures and Runtime active roots remain unproved by this report. |
| Lowering | Output identity and Runtime source spans, output ranges, callable declarations and jump destinations. These checks do not establish every C operation's meaning. |
| Verification | Whether this report has evidence of native compilation, declared execution outcomes, Python/C comparison and operation documentation coverage. |
| Capability | Requested resource-policy relationships and Runtime operation identities compared with a separately supplied registration snapshot. |
Ting feedback is advisory. It does not edit your source, change the conversion result, or enable an unsupported operation. If conversion is rejected, open the source diagnostic, correct that issue, and convert again. All six reviews still run so you can see the remaining gaps together.
| Finding | Meaning |
|---|---|
| Supported | The named check passed within the scope stated beside it. This is not whole-program approval. |
| Unknown | Evidence is absent, incomplete, unsupported or outside the inspection budget. This is not a finding that your program is wrong. |
| Contradicted | A concrete check disagrees with a recorded claim. Investigate the identified record; a mismatch alone does not locate a Python defect. |
| Conflict | Positive evidence conflicts with negative evidence or a required supporting check. |
| Not applicable | The named check does not apply to the observed inventory. |
Supporting checks must be established before dependent findings can be supported. Ting rules have traceable explanations, but an explanation of a rule is not a proof of the entire converted program. Some inspections reuse Crucible's consumer validators; they can catch inconsistent records but can share implementation mistakes with the conversion engine.
A missing check means that the review does not have the corresponding evidence. It does not by itself mean your program has failed that test. A normal workspace conversion does not compile or run generated C, so compiler checks and Python-versus-C execution comparisons remain unverified for your current source.
Use this check when you know what a converted program should do for a particular case. PyCForge builds a temporary executable and compares its actual output and exit status with the results you provide. Your saved builds are unchanged. The check covers these declared cases; it does not infer the intended answer or establish general Python/C equivalence.
The program runs with your account's file and network access. Use trusted source. Time/output limits and temporary folders are not a security sandbox. Opening the dialog, entering cases, importing JSON and ordinary conversion do not run your program.
| Field | Meaning and how to use it |
|---|---|
| Case name | A short, unique name that identifies this test in the results. |
| Program input | Text supplied to the program's standard input. Leave it empty if the program reads no input. It does not supply function arguments or command-line arguments, and is useful only when the supported converted program actually reads standard input. |
| Expected output | The text the program should print to standard output. Enter the text itself, without JSON quotes or escape codes. Empty means you expect no output. |
| Expected error output | The text expected on standard error. Leave it empty when you expect no error output; an ordinary printed message belongs in Expected output. |
| Expected exit code | An integer from 0 to 255. Zero usually means successful completion. An integer returned by the entry function becomes its exit code; returning 7 does not print 7. A supported Runtime entry returning None completes with code 0. |
| Line endings | Choose LF or CRLF separately for each text field. Press Enter where a line break is expected. No final line break is added automatically. The label below the field tells you whether it ends with a line break. |
All comparisons are exact, including spaces, capitalization and line endings. The normal form uses UTF-8 text. LF is initially selected on Linux and CRLF on Windows; select the form your program actually produces. Invalid case names, duplicate names and excessive data are explained below the form, and must be corrected before a run.
Create a Python document containing this complete program. Select Runtime conversion, then transpile it:
def main():
print("Hello from PyCForge")
return 0
Open Ting → Verify execution…, select the entry ending in .main, and enter:
| Field | Value for this program |
|---|---|
| Case name | Greeting |
| Program input | Leave empty. This program does not read input. |
| Expected output | Type Hello from PyCForge, then press Enter once. The field should say Ends with a line break, because print adds one. |
| Output line endings | Use LF for the Linux example, or CRLF for the ordinary Windows text-output route. |
| Expected error output | Leave empty. |
| Expected exit code | 0 |
Select your prepared compiler under Compiler settings and choose Compile and run. A matching run shows the greeting case as matched, the observed greeting and exit code 0. Ting's Verification tab records compilation and the declared execution comparison separately.
To see a useful mismatch, replace the expected greeting with Goodbye, retain the final line break, and run again. The result should show actual output Hello from PyCForge against expected output Goodbye. Removing the final line break also creates a mismatch. These differences do not by themselves identify a source or converter defect; in this example the expectation was deliberately changed. Restore the greeting and final line break for a matching case.
Every declared case must finish and match before the execution comparison is supported. Each case starts in a fresh empty working folder. File contents and external side effects are outside this comparison. A completed mismatch shows the observed difference without guessing its cause. A successful build followed by a timeout can support compilation while leaving execution unverified.
Standard input is limited to 64 KiB per case and each expected stream to 256 KiB; combined declarations are limited to 2 MiB. The workspace allows 120 seconds for the build, five seconds per execution and 300 seconds overall, with 256 KiB captured per stream. A timeout, cancellation, output limit or incomplete compiler record cannot establish a passing execution result.
Changing a case, compiler setting or source retires the previous result. Source changes require a new conversion; compiler or environment changes require a fresh run. The receipt binds the exact source, generated C, executable, cases and tool settings. For ordinary Python source, compile the saved C and compare its outputs with Python for representative inputs. Conversion imports need a separately reviewed reference or adapter.
The normal form requires no JSON. Choose Advanced JSON… only to exchange case files or edit exact binary data. Import JSON… validates a selected file before changing the JSON editor. Apply cases replaces the form with validated cases; Cancel keeps the existing form. Export JSON… saves the validated editor contents without applying them to the form. Existing files require an explicit replacement confirmation.
This is the greeting case with LF line endings in JSON form:
{"schema": "pycforge.execution-cases/1", "cases": [
{"name": "Greeting", "stdin": "", "stdout": "Hello from PyCForge\n", "stderr": "", "exit_code": 0}
]}
JSON uses \n for LF or \r\n for CRLF. Both expected streams and the integer exit code are required. For binary streams, use the corresponding stdin_base64, stdout_base64 or stderr_base64 field. A text field and its base64 counterpart cannot both be present. Imported binary, control-character or mixed-line-ending data that cannot be edited losslessly as ordinary text stays read-only in the form and is preserved exactly; use Advanced JSON to change it.
Headless clients can use pycforge.verification.VerificationService with immutable VerificationPlan, ExecutionCase and VerificationOptions objects. Pass the returned ExecutionEvidence to the Ting report builder in the same process. The command below accepts supplied modules through repeated --companion MODULE=PATH and Runtime conversion through --runtime-values:
python -m pycforge.verification program.py --runtime-values --cases cases.json --entry main.main --compiler-family gcc --report execution.json
The report path must be new; existing files are preserved. Exit status 0 means all declared cases matched, 1 means a completed mismatch or unsuccessful build, 2 means incomplete/invalid/stale verification, and 130 means cancellation. JSON reports retain inputs, compiler commands, identities and observed bytes for inspection, but cannot be imported as authenticated evidence. The session seal prevents accidental receipt editing or reuse from another process; it is not remote attestation or protection against code inside the same Python process. Driver identities do not inventory every system header, library or toolchain dependency. See the conversion checklist for additional independent testing.
Read the reported reason. For a resource-limit message, try a smaller self-contained example. For an installation problem, record the message and installed version. The diagnostic from conversion remains the best starting point for correcting unsupported source.
Both conversion modes produce evidence for the six reviews. The reported checks reflect that mode's conversion contract. Runtime value conversion uses checked runtime operations, validated ForgeLens emission ranges, and output/provenance checks. Ting reruns these checks against the supplied source and compares the fresh receipt with the recorded receipt; it does not treat the Typed analysis model as proof for Runtime output. A favorable review does not expand the supported source forms.
The startup splash prepares Ting's NumPy/Numba numerical routines and local review resources on a background worker. Reviews operate locally without a network connection.