Test end-to-end su un albero reale (9284218024b.asm, confrontato con la sua
BOM Creo di riferimento tramite scripts/CreoBom.cs) oltre gli 8 campioni usati
per il reverse engineering iniziale.
Trovato: i componenti creati da un pattern/array feature hanno un ConfigName
con suffisso "<codice>#AF<n>" (n = indice nel pattern), non un'istanza di
family table vera - la BOM di Creo lo ignora e unisce le occorrenze duplicate.
AsmRecordReader ora rimuove questo suffisso prima di usare il nome istanza.
Trovato ma non risolto: un intero ASSIEME (non solo le parti normalizzate)
puo' avere una family table, risolta da Creo con un meccanismo diverso da
ConfigName (non decodificato). AsmRecordReader non lo rileva e non lancia
eccezione in questo caso: il fallback automatico di Form1 non scatta, e il
risultato puo' essere silenziosamente incompleto. Documentato come limite
noto in formato-asm-record.md e in CLAUDE.md - nessun fix applicato finche'
il meccanismo non e' meglio compreso.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
AsmRecordReader (spostato da scripts/CreoBom.cs nel progetto principale,
unica fonte di verita' condivisa con lo script via #:project) sostituisce
AssemblyComponentScanner in Form1.RisolviAlberoComponenti (fase "Trova
componenti referenziati") e AssiemeCompleto (fase "Verifica completezza"):
legge esattamente la sezione "members" del binario invece di contare
occorrenze testuali, eliminando i falsi positivi/negativi noti dell'euristica
(riferimenti fantasma, placement senza suffisso, ecc.).
Se un file non rispetta il formato validato (InvalidDataException, coperto
solo dagli 8 campioni usati per il reverse engineering) Form1 ripiega
sull'euristica di AssemblyComponentScanner, che resta quindi come fallback
e come fonte di verita' per gli script BomViaDb.cs/ExtractCreoStrings.cs.
Aggiunti test per AsmRecordReader (fixture binarie costruite a mano: lettura
membri, risoluzione ConfigName, conteggio disallineato, sezione assente).
Aggiornata la documentazione (CLAUDE.md, docs/architettura.md,
docs/formato-asm.md, docs/formato-asm-record.md) di conseguenza.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Porting C# del reverse engineering della struttura a record del .asm Creo
(sezione "members" + parametro "ConfigName" per le family table): legge
esattamente i componenti diretti e genera la stessa BOM esportata da Creo,
senza le euristiche testuali (conteggio occorrenze) di AssemblyComponentScanner.
Resta uno script di analisi indipendente, non integrato nella GUI.
Aggiorna anche formato-asm.md (non e' piu' vero che "non esiste un parser
strutturale") e architettura.md (script mancante nell'elenco).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Schema, logica di scansione/upsert incrementale di Indexer.ScanAsync,
come l'app individua il DB all'avvio (incluso il caveat sulla cartella
di build in dev) e le query piu' usate.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Due bug reali trovati confrontando l'estrazione BOM con la BOM ufficiale Creo:
un componente scartato per placement fallito in un file veniva bloccato per
sempre anche quando il file giusto aveva un vero placement (perdendo interi
sotto-alberi come 9256510047), e i codici con suffisso (_C01 ecc.) richiedevano
sempre placement esplicito invece dell'eccezione per occorrenze ripetute già
presente nella GUI. Entrambi corretti.
AssemblyComponentScanner reso pubblico e scripts/BomViaDb.cs, ExtractCreoStrings.cs
riscritti per referenziarlo via "#:project" invece di duplicare l'euristica:
un'unica fonte di verità condivisa da GUI e script standalone.
Aggiunti test di integrazione per Indexer.ScanAsync (mai coperto oltre al
parsing del nome file) e test end-to-end che ricostruiscono l'intera pipeline
disco -> indice -> risoluzione albero componenti senza toccare la UI.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Not needed anymore: the app no longer copies matched .stp files from
step/ into a separate step_finali/ folder, or reports which distinta
codes are still missing a STEP. Removed the "Crea cartella step_finali"
button, GeneraReportStep, and ReportStepForm/EsitoStep. Phase 5 keeps
"Verifica completezza assiemi" and the drag&drop .asm list (still how
files get sent to Creo Distributed Batch); renamed from "Verifica
completezza e STEP" to "Verifica completezza" since it's no longer about
STEP output. PulisciRawEStep now only clears raw/ and step/.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Working folder (phase 1): CSV/raw/step/step_finali all used to live split
across two different implicit locations (CSV in the "repo root" found by
walking up for CLAUDE.md, raw/step next to the input distinta file) with
neither shown to the user. That repo-root lookup also silently breaks once
the app runs as a standalone published exe on a machine with no repo/
CLAUDE.md at all. Replaced both with one explicit "cartella di lavoro",
auto-suggested (last used, remembered the same way as the DB path) and
changeable from phase 1, with no dependency on any repo marker file.
Rename: CadSearchGui was the working name from before the app had a
product identity; renamed folders, csproj files, namespace and the
AppData settings folder to PrepStep throughout, matching the exe name
already in use. Product-facing name stays "Cevolani StepPrep".
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- New app.ico (from the Cevolani favicon, background made transparent and
cropped tighter to the logo so it stays legible at 16/24px), wired up via
ApplicationIcon in the csproj and re-read at runtime for the window icon.
- Named the product "Cevolani StepPrep" and set Version=0.1.0 in the csproj;
shown in the window title. The CadSearchGui folder/namespace/csproj keep
their technical name, only the product-facing name changed.
- Updated CLAUDE.md, README.md and docs/architettura.md to reflect the
product name/version, UiStile.cs, app.ico, and that this app is a
pre-process step ahead of Creo Distributed Batch.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
CopiaFileTrovati now copies with Parallel.ForEach (degree 6) instead of
sequentially: the bottleneck is network latency to the share, not CPU,
so overlapping copies gives a large speedup (~5x in a synthetic
benchmark with simulated SMB latency, documented in
docs/copia-raw-parallela.md).
Also adds docs/architettura.md, a full architecture overview of every
source file, the wizard's 5 phases, dialogs, persistence and standalone
scripts.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Pro-E numeric codes cover most components, but parts downloaded as STEP
from online catalogs (rings, bushings, knobs...) get alphabetic Creo names
with no predictable shape. Add a second extraction path that tokenizes the
binary generically and validates candidates against the known codes in
cad_index.db, applying the same placement/occurrence rule used for bare
numeric codes. Numeric extraction is unchanged, so missing-from-index
reporting for renamed/moved Pro-E codes still works.
Also load all known prt/asm codes once per scan instead of one query per
candidate, in both RisolviAlberoComponenti and VerificaCompletezzaAssiemi.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Two more real placement forms found by comparing against Creo BOM reports:
- A sub-assembly can be referenced only through a "family_name\0<code>\0"
metadata field instead of a textual "Componente ID" placement or the
doubled-NUL pattern (case 9284202452 inside 9284208188_D400.asm, which
contains 9284202452_C01/_C02/_C03 - previously dropped entirely, along
with its three children).
- A bare numeric code immediately followed by a family-table bracket
(<code><GENERIC>, e.g. bearing instances like 2510060303<CUSCI_ACBB7>)
is a real reference to that instance, even though only the generic file
is indexed on disk. Previously the instance code silently vanished before
ever reaching the DB lookup; now it survives placement filtering and ends
up in the "scartati" (discarded) list instead, so it's traceable rather
than invisible.
docs/formato-asm.md documents both, bringing the total to four recognized
placement forms.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The bare root code of an assembly (e.g. 9284208188, without a _D... suffix)
is always referenced inside its own suffixed sub-assemblies as a genuine
placement pointing back to the general assembly. Following it during
recursive resolution pulled in the entire product tree instead of just the
requested sub-assembly (598 components collected instead of ~65 on
9284208188_D600). Fixed by seeding the visited-codes set with each starting
assembly's root code, in both scripts/ExtractCreoStrings.cs and the two BFS
call sites in Form1.cs (RisolviAlberoComponenti, AssiemeCompleto).
Also closes two recall gaps found by comparing against a real Creo BOM
report: some real placements only show up as the code repeated twice
separated by a NUL byte rather than the usual "Componente ID N (...)" text,
and family-table bearing codes (CUSCI_...) use the same <...> bracket
convention as UNI hardware but weren't recognized.
docs/formato-asm.md documents all three findings.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- Batch the DB lookup per file instead of per code in scripts/ExtractCreoStrings.cs:
files_ultima_versione is an unmaterialized view, so each query rescanned the whole
files table (~230ms) - batching cuts a multi-minute BFS to seconds.
- Widen the code regex to accept a revision letter and a "_..." suffix together
(e.g. "9284200366A_C01"), and purely numeric "_NNNN" suffixes (e.g. dimensional
variants like "2702220182_8188"), not just letter+digits.
- Require bare digit-only candidates to also appear inside a real
"Componente ID N (...)"/"Component id N (...)" placement, filtering out
incidental references to a containing assembly's own code.
- scripts/ExtractCreoStrings.cs now reports only the direct references of the given
.asm (one level, matching Creo's own tree for that file) instead of recursing into
sub-assemblies.
- Port the regex and placement-check refinements to the production
AssemblyComponentScanner.cs used by the app's "Trova componenti referenziati" and
"Verifica completezza" phases.
- Document the new false-positive categories (parent/container self-reference,
suppressed/removed components with residual placement, placement missing
suffix/bracket detail) in docs/formato-asm.md, generalized beyond the specific
assemblies used to find them.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
These appear in the .asm binary as a family-table instance name (e.g.
"R6--_-_U6592<UNI6592>"), not a separate file on disk - only the generic
"uni6592.prt" is actually indexed. Extract just the "<UNI...>" content
instead of the full instance name, in both the standalone script and the
production scanner, so standard hardware stops being silently dropped as
"not found in index".
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Reverse-engineered the Creo .asm binary format enough to extract referenced
component codes via a text heuristic, mirroring AssemblyComponentScanner's
approach and Form1's recursive resolution against cad_index.db. Documents
known false-positive classes (ModelView# view IDs, self-references, and
phantom references to superseded/renamed codes).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
davideandClaude Sonnet 5