One user-space repair task per eligible HWE debug-diag / compile-deploy base ID: 320 admitted (C 218, C++ 97, build scripts 5), 1 skipped by the content filter. Each ships a public spec with a check-to-basis table, at least two defects with single-fix fixtures, a symptom-masking fixture and an alternative correct implementation. Claude re-checked all 320: structure, the HWE quality items and trace6 Docker calibration receipts against current digests (320/320 USABLE). Difficulty is unmeasured. Known follow-up: 212 tasks use 120 ad-hoc family names; the next batch maps them to the canonical list. The batch's 1,256-line progress log stays in its factory report and the 36-local branch batch15-hwe. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
7.1 KiB
Array report contract v1
C1 Build and interface
Use make with the available native gcc C11 toolchain. This is software execution on x86_64, not ARM certification.
The four submitted translation units link with array_report.h and observer.c; no external dependencies or extra files are used.
The callable entry is int ra_process(const ra_request *, const ra_port *); the header defines all structures and helper interfaces.
All helper functions declared by the header remain linkable; their outputs have the definitions below.
A port offers allocate(context, bytes), release(context, pointer), and publish(context, report).
Each invocation is independent. Input structures must not be changed. stderr may contain arbitrary diagnostic bytes without a size limit.
C2 Validation and results
A request is valid exactly when count=0..128, seed=0..65535, base=-100..100, step=-7..7, mode=0..3, cutoff=0..100, window=1..16, bins=1..8 and fail_at=0..2. Missing request/port/callback is invalid. Invalid requests return -2 without allocating, releasing or publishing. A valid count=0 returns 1 with no callbacks. For nonempty valid input request countsizeof(ra_cell) bytes, then countsizeof(unsigned) bytes, in that order, once each. The allocator returns NULL on call fail_at when fail_at is 1 or 2. Otherwise it returns distinct malloc-backed storage. A NULL result returns -1 without publishing. Release every successful allocation exactly once on every exit path; never release NULL or unknown storage. On success publish exactly one fully computed report, release both buffers and return 0. Publication occurs before either buffer is released. No failed invocation publishes partial output. No unbounded allocation, real OOM, overflow or freed-storage access is needed.
C3 Source array and selection
Begin with uint32_t word=seed. At each original position i=0..count-1, replace word by (1664525*word+1013904223) modulo 2^32.
Let raw=base+(i modulo 11)*step+((word>>16) modulo 101)-50.
mode 0 keeps raw; mode 1 takes abs(raw); mode 2 clamps raw into [-25,25]; mode 3 takes sign(raw)rawraw (zero stays zero).
Keep a transformed cell exactly when abs(value)>=cutoff. Keep original positions and preserve source order in the selected sequence.
Only the first selected_count cells and order entries are defined; unused tails have no required contents.
ra_valid, ra_transform and ra_generate implement these definitions.
C4 Report fields 0..22: sequence analytics
All 56 fields are signed long long; initialise them to zero and then set the fields below. All results fit signed 64 bits.
0 original count; 1 selected count n; 2 rejected count; 3 sum; 4 sum of absolute values; 5 sum of value squares.
6 sum of (original_position+1)*value. 7 minimum; 8 maximum; 9 original position of first minimum; 10 original position of first maximum.
11 negative count; 12 zero count; 13 positive count; 14 sum of even values; 15 sum of odd values.
16 longest contiguous nondecreasing run length; 17 longest contiguous nonincreasing run length; empty sequences have run length zero.
18 sum of abs(difference) between neighbours of the selected sequence.
19 largest sum of a window of min(window,n) consecutive selected cells; 20 first selected-sequence index attaining it.
21 largest sum of a nonempty contiguous segment (may be negative). For empty sequences fields 7..10,19..21 remain zero.
22 uint32 checksum: begin 2166136261, for each selected cell xor uint32(value), multiply 16777619 modulo 2^32,
then xor original_position and multiply 16777619 modulo 2^32. Empty checksum is 2166136261.
ra_analyse sets exactly fields 3..22.
C5 Report fields 23..35: order analytics
ra_order writes indexes of selected cells sorted by value ascending, ties by original position ascending.
23 number of distinct values; 24,25,26 sorted values at floor((n-1)*k/4) for k=1,2,3 respectively.
27 sum of (sorted_rank+1)*sorted_value; 28 largest gap between sorted neighbours (zero for n<2).
29 sum of abs(value-lower_median), where lower_median is field25. 30 original position of first sorted cell.
Partition [minimum,maximum] into bins integer buckets: index=floor((value-minimum)*bins/(maximum-minimum+1)).
When minimum=maximum every selected cell has bucket0. 31 earliest bucket with largest count; 32 that count;
33 number of occupied buckets; 34 sum of (bucket+1)*bucket_count. All fields23..34 are zero for empty selection.
35 number of pairs i<j in selected source order with value[i]>value[j]. ra_rank sets exactly fields23..35.
C6 Report fields 36..55: signal analytics
36 minimum prefix sum and 37 maximum prefix sum, including the empty prefix of sum0.
38 number of strictly positive contiguous runs; 39 number of strictly negative contiguous runs; zero terminates both kinds.
40 longest strictly positive run length; 41 longest strictly negative run length.
42 number of strict interior local maxima; 43 number of strict interior local minima; endpoints are excluded.
44 number of equal neighbour pairs; 45 sum of products of neighbours.
46 largest count of skipped original positions between selected neighbours; 47 total skipped positions between selected neighbours.
48 length of longest strictly increasing subsequence; 49 length of longest strictly decreasing subsequence (not necessarily contiguous).
50 most frequent value, tied by its first occurrence in selected source order; 51 its frequency; 52 number of equal-value pairs i<j.
53 alternating sum value0-value1+value2-... in selected source order. Empty sequence fields36..55 are zero.
54 number of sign changes after removing zero values; 55 longest contiguous equal-value run length (empty=0).
ra_trends sets exactly fields36..55.
C7 Public observer protocol
observer reads one decimal request per stdin line, nine fields in header order, until EOF; public examples use decimal ints. Counts -2,-3,-4,-5,-6 probe missing request, port, allocate, release, publish respectively, with an otherwise count=1 request. Its callback instrumentation records allocation calls, release calls, live allocations, invalid releases, two requested byte sizes, and publish calls. Release removes a tracked allocation; invalid release increments invalid_releases. After observing a request, observer safely frees any leftover tracked allocation. The publication observer copies the 56 report fields and records a lifecycle violation as invalid_releases if either required allocation is no longer live. Any change to request or port fields also increments invalid_releases. stdout has exactly one line per request: return_code allocation_calls release_calls live_allocations invalid_releases size1 size2 publish_calls then 56 report fields. Before a publication the report copy is zero. Only the defined report fields are observed, never unused tails. Each token is signed decimal ASCII: optional + or -, arbitrary leading zeros, separated by one or more SPACE/TAB/CR bytes. One LF separates records, with optional final LF. No extra records or other stdout text are allowed. stderr is diagnostic only. observer's successful exit is 0. An input syntax error is 2. Required grading inputs are within C int range, including explicit invalid-contract requests.