Files
simple/docs/test-scripts.md
T
catarrh 0b9e9e4c5e v3.17.0: test scripts for the applet handoff - menu by text, spec-order TRs, codings
The applet-testing batch agreed after v3.15/3.16:

- menu-select by text: the step names the item by `text` (+ `mode`
  exact/contains, `case_sensitive`) instead of the install-dependent id
  (both or neither is refused); the runner resolves it against the card's
  cached menu and raises a step error naming the available items when nothing
  (or several items) matches.  The cache is refreshed on every SET UP MENU
  fetch - the proactive chain and the expect step - so an applet install (new
  items/ids after its REFRESH) no longer leaves a stale menu behind.  The
  step editor offers the id or the text + mode and the client-side pre-check
  bounds them.
- TERMINAL RESPONSE object order: both builders emitted the Item identifier
  before the Result (and the scripted builder its Text string before it too);
  TS 102 223 6.8.0 mandates Command details, Device identities, Result,
  Duration, Text string, Item identifier.  `testscript.build_tr` and
  `_menu_send_response` now follow it; `respond.raw` is documented as
  appended after the standard objects, and the exact-byte tests pin the order.
- Text codings: the scripted response codes the GSM default alphabet via the
  GSM 03.38 codec ('00' packed, '04' 8-bit) instead of latin-1, keeps '08'
  UCS2 and sends the spec's null text string for an empty answer; the packing
  helper is shared with the interactive path (`testscript.pack_gsm7`).  A
  UCS2 GET INPUT answer is documented and tested.
- docs/test-scripts.md gained the handoff brief for the applet agent: the
  value-collection table, fill-in worksheets for the two applet scenarios, a
  menu-by-text example, the validation checklist and a format stamp; help
  EN/RU and docs/api.md updated.

Tests: 809 frontend + 705 Python; sw cache simple-v380.
2026-10-04 17:48:29 +03:00

14 KiB
Raw Blame History

Test script authoring guide

A test script is a JSON document that drives a deterministic dialogue with the card over the local server: ordered actions (each sends something and checks the response) and expectations (each fetches the proactive command the card announced, checks it, and sends the scripted TERMINAL RESPONSE).

Scripts live in the server-side store (~/.pysim-simple-server/test_scripts.json, --test-scripts): the PWA's Test script pill manages them (Import/Export move JSON in and out, the old localStorage export imports as-is), and any API client can create or run them (POST /api/test/run with an inline script or a stored script_id - see docs/api.md). While a run is active the card belongs to the script.

Format: server ≥ 3.15.0 (kvn, por and files checks), scripts stored server-side since 3.16.0, menu-select by text and the spec-order TERMINAL RESPONSE codings since 3.17.0.

Writing a test for an applet

Collect these values before writing the JSON (from the applet's source/spec and the card's preset):

What Where it goes Notes
Applet TAR (3 bytes) + MSL, or an explicit SPI1 scp80.params.tar / spi1 the card preset must carry the TAR with its MSL, or the step passes spi1
Keyset number scp80.params.kvn (1–15) only when the applet's keys are not the preset's first keyset
Trigger C-APDUs scp80.params.apdu one secured packet per scp80 step; the preset provides keys and counter
Expected file paths + contents file-read.params.path + check.data a mask must have exactly the actual value's length
Expected proactive command(s) expect.command + qualifier/checks REFRESH qualifier: "00" = NAA init + full FCN, files asserts the changed EFs
PoR transport + expected PoR scp80.check.por/check.data (inline) or expect SEND SHORT MESSAGE + por check pick per the SPI2 the applet expects
Delivery timing a status action when the card delivers on poll the runner never polls by itself

Worksheet 1 - RFM update → file change → REFRESH (full FCN)

{"name": "applet RFM update",
 "steps": [
  {"type": "action", "kind": "scp80",
   "params": {"apdu": "<RFM C-APDU>", "tar": "<applet TAR>", "kvn": 1, "spi2": "01"},
   "check": {"sw": {"mode": "mask", "value": "91??"}, "por": "none"},
   "label": "RFM update"},
  {"type": "expect", "command": "REFRESH", "qualifier": "00",
   "checks": [{"kind": "files", "value": "<changed FIDs, e.g. 3F007F206F07>"}],
   "respond": {"result": "ok"}},
  {"type": "action", "kind": "file-read", "params": {"path": "<EF path>"},
   "check": {"data": {"mode": "mask", "value": "<expected bytes, ? = wildcard>"}}}
 ]}

Worksheet 2 - incoming data → particular PoR, no other actions

Inline PoR:

{"name": "applet data (inline PoR)",
 "steps": [
  {"type": "action", "kind": "scp80",
   "params": {"apdu": "<data C-APDU>", "tar": "<applet TAR>", "spi2": "01"},
   "check": {"sw": "9000", "por": "ok",
             "data": {"mode": "mask", "value": "<expected PoR bytes>"}}}
 ]}

PoR-in-submit:

{"name": "applet data (SEND SM PoR)",
 "steps": [
  {"type": "action", "kind": "scp80",
   "params": {"apdu": "<data C-APDU>", "tar": "<applet TAR>", "spi2": "21"},
   "check": {"sw": {"mode": "mask", "value": "91??"}, "por": "none"}},
  {"type": "expect", "command": "SEND SHORT MESSAGE",
   "checks": [{"kind": "por", "status": "por_ok", "sw": "<R-APDU SW>",
               "data": {"mode": "mask", "value": "<expected response data>"}}],
   "respond": {"result": "ok"}}
 ]}

Validation checklist

  1. Client-side: the PWA import (or testScriptProblem) rejects structural mistakes; the server validates again with the same engine at run start.
  2. Import the JSON (PWA → Simulator → Test script → Import, or POST /api/test/scripts), select the matched preset, Run.
  3. Read the report: every action/expectation shows expected vs actual per check; ⚠ = warning (continues), ✗ = error (stops). A failed step names the reason (note).
  4. Iterate: a mismatch on data usually means the mask length differs from the actual value, or the value changed between runs - mask the varying bytes (counter, date-time) with ?.
  5. Counters: the run starts from the preset's stored counter and persists each accepted packet's next value; consecutive runs continue correctly.

Document shape

{"name": "applet RFM update",
 "steps": [
   {"type": "action", "kind": "scp80", "params": {…}, "check": {…}, "label": "optional"},
   {"type": "expect", "command": "REFRESH", "qualifier": "00",
    "checks": [{"kind": "files", "value": "3F007F206F07"}],
    "respond": {"result": "ok"}}
 ]}

Every step may carry on_fail: "error" (default - stops the run) or "warning" (continues, the report shows ⚠). Each entry of checks may override it with its own on_fail.

Presets: where the keys, TARs and counters come from

An SCP80 action never carries key material. The PWA sends the stored preset's id with the run; the server resolves it and takes:

  • the keyset - the step's kvn (1–15) or the preset's first keyset,
  • the keys and the counter of that keyset,
  • the TAR table with each TAR's MSL (Minimum SPI1).

So before a run the matched preset (ICCID) must contain the keyset and, for every TAR the script addresses, either a TAR entry with its MSL or an explicit spi1 in the step. Counter handling: the run starts from the keyset's stored counter and persists each accepted packet's next value, so consecutive runs continue correctly.

Actions

{"type": "action", "kind": "…", "params": {…}, "check": {…}}
kind params notes
envelope event 0–255, data hex (optional) ENVELOPE(Event Download)
menu-select item_id 1–255 or text + mode exact/contains (+ case_sensitive), not both ENVELOPE(Menu Selection); the text is resolved against the card's cached menu - refreshed whenever the card sends SET UP MENU, so it follows the applet's install parameters
file-write path, data, mode auto/binary/record, record 1–255 UPDATE BINARY/RECORD
file-read path, mode, record READ BINARY/RECORD; verify with check.data
apdu apdu hex raw transport, no auto-handler
scp80 apdu xor sp (optional source), kvn, tar, spi1, spi2 secured packet to the TAR
status attempts 1–1000, interval_ms 0–10000 with attempts > 1 the default SW check is 91?? (poll until a command)

path is /-separated: MF (or 3F00) or an ADF name/AID first, then FIDs or file names - e.g. MF/7F20/6F07, ADF.USIM/EF.TEST.

check on actions

"check": {"sw": {"mode": "mask", "value": "91??"},
          "data": {"mode": "exact", "value": "0271…"},
          "por": "ok"}
  • sw - default {"mode": "exact", "value": "9000"} (a status action with attempts > 1 defaults to the 91?? mask).
  • data - the response data (for SCP80: the inline PoR packet).
  • por - "any" (default, no check), "ok" (response_status == por_ok), "none" (no PoR expected).
  • A plain hex string is exact; a value containing ? is a mask. A mask must have exactly the same length as the actual value (? is a per-nibble wildcard) - e.g. 91?? matches 9102, 027100???? matches a 4-byte prefix.

Expectations

{"type": "expect", "command": "SEND SHORT MESSAGE",
 "qualifier": "00",
 "checks": [{"kind": "por", "status": "por_ok", "sw": "9000"}],
 "respond": {"result": "ok"}}

The card must have announced a command (91XX) in the previous step - the runner never polls by itself. command is a name (as shown in the proactive log) or a hex type; ANY/* skips the type check. qualifier is the command-details byte 3 (one byte, exact/mask).

Content checks

kind fields checks
text mode contains/exact, value, case_sensitive alpha identifier / text
item id, text, mode, case_sensitive items of SELECT ITEM / SET UP MENU
raw mode exact/mask, value the whole fetched command (hex, exact length needed)
por status, sw, data the PoR carried by a SEND SHORT MESSAGE (see below)
files files (list of path hex strings) or a comma/space separated string the File List of a REFRESH, TS 102 223 §8.18; matched order-insensitively

por needs an SCP80 step before the expectation: the TPDU of the SEND SHORT MESSAGE is decoded with that step's keyset, SPI and counter. status is the decoded response status name (por_ok, cntr_low, rc_cc_ds_failed, …; ok is accepted for por_ok); sw is the R-APDU status word and data the R-APDU response data (exact/mask) - both come from the decoded response (scripting AB/AF/compact forms). A PoR split over several SMS parts is accumulated across SEND SHORT MESSAGE expectations: put the por check on the step that completes the sequence (intermediate steps should use raw or no PoR check).

respond

The TERMINAL RESPONSE for the fetched command, in the TS 102 223 §6.8.0 object order (Command details, Device identities, Result, Duration, Text string, Item identifier, …):

  • result - name or hex (ok, partial, missing, refused, not_understood, modified, cancel, back, timeout, no_response).
  • text + dcs - the Text string for a GET INKEY / GET INPUT answer. dcs uses the spec's recommended codings: "00" GSM default alphabet 7 bits packed, "04" GSM default alphabet 8 bits (one octet per character), "08" UCS2 (UTF-16-BE). An empty text sends the null text string (Length 00).
  • item_id - the Item identifier for a successful SELECT ITEM answer.
  • raw - extra COMPREHENSION-TLVs appended after the standard objects.

UCS2 answer to a GET INPUT:

{"type": "expect", "command": "GET INPUT",
 "respond": {"result": "ok", "text": "Пароль", "dcs": "08"}}

The two PoR transports (SPI2)

  1. Inline - the card answers the ENVELOPE 9000 with the PoR in the response data. Verify on the action: "por": "ok" and/or "data": {"mode": "mask", "value": "…"}.
  2. PoR-in-submit (SPI2 bit 20, e.g. spi2: "21") - the card announces a SEND SHORT MESSAGE (91XX); the PoR is inside its SMS TPDU:
{"type": "action", "kind": "scp80",
 "params": {"apdu": "…", "tar": "B00001", "spi2": "21"},
 "check": {"sw": {"mode": "mask", "value": "91??"}, "por": "none"}},
{"type": "expect", "command": "SEND SHORT MESSAGE",
 "checks": [{"kind": "por", "status": "por_ok", "sw": "9000"}],
 "respond": {"result": "ok"}}

If the card announces the SEND SHORT MESSAGE only on a later STATUS poll, insert a status action (attempts > 1) before the expectation.

"No actions" after a trigger

An exact SW check is the assertion: "check": {"sw": "9000"} fails if the card answers 91XX (a command was announced); the runner also errors on a pending command the next step does not expect.

Worked examples

Menu selection by text (the id varies with the install parameters)

{"name": "applet menu",
 "steps": [
  {"type": "action", "kind": "menu-select",
   "params": {"text": "Settings", "mode": "exact"}},
  {"type": "expect", "command": "ANY", "respond": {"result": "ok"}}
 ]}

The text is matched against the card's cached menu; the cache is refreshed whenever the card sends SET UP MENU (e.g. after an install + REFRESH). If the script must wait for the menu, add a status action and an expect SET UP MENU before the selection.

RFM update → file change → REFRESH (NAA init + full FCN)

{"name": "applet RFM update",
 "steps": [
  {"type": "action", "kind": "scp80",
   "params": {"apdu": "80E2900000", "tar": "B00001", "spi2": "01"},
   "check": {"sw": {"mode": "mask", "value": "91??"}, "por": "none"},
   "label": "RFM update"},
  {"type": "expect", "command": "REFRESH", "qualifier": "00",
   "checks": [{"kind": "files", "value": "3F007F206F07"}],
   "respond": {"result": "ok"}},
  {"type": "action", "kind": "file-read", "params": {"path": "MF/7F20/6F07"},
   "check": {"data": {"mode": "mask", "value": "AA??CC"}}}
 ]}

qualifier 00 is "NAA Initialization and Full File Change Notification"; the files check asserts the exact set of changed files the applet reports. For this mode the terminal must not reset the card (TS 102 223 §6.4.7), so the run continues - the following file-read verifies the updated content.

Incoming data → particular PoR, no other actions

Inline PoR:

{"name": "applet data (inline PoR)",
 "steps": [
  {"type": "action", "kind": "scp80",
   "params": {"apdu": "80E2900000", "tar": "B00001", "spi2": "01"},
   "check": {"sw": "9000", "por": "ok",
             "data": {"mode": "mask", "value": "027100????0102"}}}
 ]}

PoR-in-submit (with the transport asserted by the expectation):

{"name": "applet data (SEND SM PoR)",
 "steps": [
  {"type": "action", "kind": "scp80",
   "params": {"apdu": "80E2900000", "tar": "B00001", "spi2": "21"},
   "check": {"sw": {"mode": "mask", "value": "91??"}, "por": "none"}},
  {"type": "expect", "command": "SEND SHORT MESSAGE",
   "checks": [{"kind": "por", "status": "por_ok"}],
   "respond": {"result": "ok"}}
 ]}

Run errors worth knowing

  • SCP80 preset is incomplete: … - the matched preset has no complete keyset; step N: keyset K is not defined in the preset - the step's kvn is outside the preset's keysets.
  • no MSL for TAR … - set it in the Cards tab or pass spi1 - the step's TAR has no entry in the preset and the step carries no spi1.
  • no proactive command pending … - an expectation ran without a 91XX from the previous step (add a status action when the card delivers on poll).
  • card verdict: no PoR (the card sent none for SPI2 …) - an SCP80 operation expected a PoR that never arrived (check SPI2 and the transport).

Known limitations

  • The runner executes one command at a time and never polls by itself; a proactive command that arrives only on a STATUS poll needs an explicit status action.
  • A por check only decodes SEND SHORT MESSAGE commands; the inline transport is checked on the action (por/data).
  • files matches the File List as a set (order-insensitive); duplicate paths are kept.