Skip to content

Version 0.1.0 is a first version, for trial. See what is not proven →

bogolang
FR

← Documentation

Not proven

What was never tried on real equipment, said plainly.

All documentation pages

This page lists, plainly, what has never been tried on real equipment or on other machines. Every part of Bogolang has automatic tests, but most tests run on saved answers written for them, or on the machine itself (127.0.0.1). A test proves the code does what its writer expected; only a trial proves that a device answers as expected. Read this page before you trust a command on a network that matters.

Version 0.1.0 is a first version for trial.

What was tried on real equipment

So that the lists below read in proportion, this is what has met the real thing:

  • One Cisco IOS-XE 17.8.1 switch: show version through a session, SNMPv3 gets and a walk (SHA and AES), two snapshots compared, a route lookup, log events and CDP neighbours read.
  • One real change, on one IOS-XE 03.06 access switch: a description on an unused port, with a verify that was false on purpose (the rollback was sent), then an add and a remove.
  • One Meraki organisation read through the Dashboard API, and one inventory read from the API of a monitoring system.
  • Junos 26.2 on virtual switches: read-only collection, the readers, and a change block with commit confirmed.
  • ping, trace and "do not fragment" to a public address, from Windows and from Linux, with routers on the way answering; DSCP marks checked on arrival.

Devices and sessions

  • Platforms never met for real: NX-OS, AireOS, ASA, IOS 15, and Junos on real hardware. Their outputs in the tests were written by hand.
  • Not tried on a device: the enable step, .table() on a live answer, a parallel loop over several real devices.
  • The jump host (ssh(d, via=jump)) has not met a real jump server.
  • The serial console has not met a real console port.
  • RESTCONF has not met a real device. NETCONF is not built.
  • optics has not read a real transceiver, on IOS or on Junos. Its reader was written from the known form of the command; another release may word its table differently and give no readings.
  • snapshot and compare have not found a real change on a device, nor read an ASA, a device with CDP turned off, or a router with VRFs (the plain command reads the global table only).
  • SNMP version 2c has not been tried on a device.
  • The configuration object and the audit rules were tested on configurations written for the tests. Watch for a banner whose text holds its own end mark, long VLAN lists wrapped over several lines, and the layouts of ASA and NX-OS.
  • No program in the class form has run against a device. The thirteen examples ran on saved answers written by hand.

Changes

  • revert_after: the device timer commands have never been accepted by a real IOS device. Check them on the exact release before you trust them.
  • Not seen for real: a device that refuses a line, a lost session during a change, Ctrl+C on a real device, a change on ASA or NX-OS.
  • When a device refuses one line of several, the run record marks all the apply lines as failed. It does not show which went in before the refused one. The rollback is sent in every case.
  • A warning from the device that is not a refusal is not noticed. The verify lines are the real test.
  • On IOS the apply lines are sent with no test of their first word.
  • On Junos: one release, on a virtual switch only. The 30 second guard, a lost session and Ctrl+C were tried in the tests only.
  • The rules stop accidents. They are not a security wall. The real limit is what the device account may do.

Reach

Not proven, since each needs a real path:

  • a trace over IPv6 beyond ::1;
  • a router that answers "cannot be reached";
  • answers from another machine through the Windows firewall;
  • a DSCP-marked ping on Windows on a PC other than the one it was tried on (it needs a raw socket, which Windows may refuse to an ordinary user);
  • "do not fragment" on Windows beyond the fact that Windows accepts it (Windows does not say the limit of the path);
  • the Linux installer step for the ping setting with a real person at the keyboard. It was tried with a tool typing the answers. The path where a container takes only the narrower value was not run as a whole.

Logs, listeners and watching

  • No device has sent a syslog message, a trap or a NetFlow record to Bogolang. The packets of the tests follow the published forms; real makers bend them.
  • No program has listened on port 514 or 162; the tests use high ports.
  • The speed of NetFlow under a busy exporter is not measured.
  • watch has not drawn on a real terminal (the Windows console, Windows Terminal). Ctrl+C on a listening or watching program was not pressed on a real keyboard.
  • Log events, route lookups and topology have read one real device. Other releases may show forms the tests do not have, above all the time stamps of logs and the port names that LLDP neighbours announce.
  • No capture file made by a real tool has been read. The test files were built from the format documents.

Mail, reports and other services

  • mail has not talked to a real mail server.
  • No PDF report has been made. The conversion needs Microsoft Word or LibreOffice.
  • No webhook has been posted to a real service.
  • The service checks (dns_check, ntp_check, http_check, cert_check, sip_check) have not met a real server, a real certificate store or a real SIP device.
  • mos is an estimate from a simplified model, not a measurement.

Installation and tools

  • A PC with no Python at all. Both installers were proven on machines that have other Pythons, which is the harder case for isolation. A clean PC was not tried.
  • The pages of the Windows installer as a person sees them. The installer was run without its windows. The menu entries were read and their commands run, not clicked.
  • A Linux of the Red Hat family. Only Ubuntu and Debian were tried.
  • A Linux older than the test systems. The carried objdump and objcopy need nothing of the system, but the engine needs its C library.
  • The Linux installer on a machine with no network, and its menu entry on a real desktop.
  • The examples and these pages inside a built installer. The build carries them; nobody has opened them in an installed copy.
  • bogo shell at a real keyboard. It was tried with lines fed to it: line editing and Ctrl+C on a running line are not proven.
  • bogo pack: a packed program that opens a real session; --login typed at a keyboard inside a packed file; a packed project with packages added by bogo add; a packed file with listeners, watch, a report or a chart; antivirus products other than the one of Windows.
  • A project on another engine than the one Bogolang carries, in the installed layout.
  • bogo pack and the own-resources code have no independent tests, written by someone other than their writer.
  • The size of the Linux installation (about 444 MB, against about 264 MB on Windows) was not looked into.
  • Python 3.15: the list of new syntax that was tried (lazy import, unpacking in comprehensions, frozendict) was not compared with the full "What's new in Python 3.15".

Limits that are known, not faults

These are not gaps in the trials; they are how Bogolang works today:

  • The access-list test reads the list, not the device: it does not know where a list is applied or in which direction. It knows nothing of ASA lists, IPv6 lists and object groups.
  • The listeners take IPv4 and UDP only, no version 3 trap, no sFlow.
  • == between two different units is False, not refused before the run (types-and-units.md).