GitLab

NetTime intercepts OS_Word 15, but when calling RTC_Adjust was not disabling that interception, so the time got set twice, ending up back at a copy of the (incorrect) soft clock.

Version 0.43. Tagged as 'NetTime-0_43'

When the RTC module is present don't try making the duplicate adjustments in this module, while largely harmless the status report would likely be out of sync as the two modules battle eachother for who can poke timer 0 more often.
Changed files
 machine.c: double quotes in debug text removed
 ntp.c: use some named defines for OS_Hardware calls
 Spec: typo correction
 rtcadjust.c: when RTC module is active, delagate adjustments to RTC_Adjust SWI

Tested on a Risc PC over 24 hours.

Version 0.42. Tagged as 'NetTime-0_42'

Rather than requiring a unary minus in rtcadjust, simply work out the offset the other way round (+ve means the RISC OS soft clock is fast compared with real time).

Version 0.41. Tagged as 'NetTime-0_41'

No longer trampled on by the current state, therefore reports the correct last state when the link is down. Also marks the last set as unsuccessful on every DNS lookup just incase someone kills the resolver too.

Version 0.40. Tagged as 'NetTime-0_40'

Retagged as NetTime-0_39.

Retagged.

Shortly to be handled by TerritoryManager, but the DST fiddling in this module was already broken. Aside from only knowing about Europe and Australia, the Australia rules were wrong too (since they've been changed again).
dhcp.c: use names for timezone CMOS rather than magic numbers.
Rework read_os_var() to remove the 'read length when buff = NULL' which is never used, but also wouldn't have worked because the read length call to OS_ReadVarVal returns an error which is caught by the _swix() hence you always got -1.
Put resolver variable name into a define in the header file.

Version 0.39. Tagged as 'NetTime-0_39'

SWI numbers and some bit fields for both C and assembler use.

Version 0.38. Tagged as 'NetTime-0_38'

Makefile changed to use "CModule" fragment.
Soak up some compiler warnings.
Extralong functions removed.
Use central header files in preference to local defines.
A failure to register resources no longer leaves EventV and WordV claimed.

Version 0.37. Tagged as 'NetTime-0_37'

Detail:
Admin:  no other changes



Version 0.36. Tagged as 'NetTime-0_36'

Detail:
  We now simply keep a flag indicating if we have ever seen the configured
  NTP server with primary sync. While the flag is FALSE, we will accept NTP
  responses more flexibly. For example, we ignore leap indicator of 3 or
  stratum of zero or greater than 14.
Admin:
  Works as hoped with the servers at Customer A.

Version 0.35. Tagged as 'NetTime-0_35'

Detail:
  NetTime will accept NTP responses from servers which do not have
  primary sync until and after they gain it. It will not accept the
  response from a server whos stratum increases, for example when
  it loses primary sync.
Admin:
  Tested on Iyonix and works as desired without breaking existing
  behaviour.

Version 0.34. Tagged as 'NetTime-0_34'

Negative timezone offsets received from DHCP were incorrectly rounded, resulting in -5:45 being set instead of -6:00.

Version 0.33. Tagged as 'NetTime-0_33'

* Improved *NetTime_PollInterval to make sure next poll happens at
  the expected time.
* On module init, no longer sets the time from the NVRAM if the current
  time seems sensible (ie >= 2003ish).
* Sets the time NVRAM on shutdown (was only on packet reception).

Version 0.32. Tagged as 'NetTime-0_32'

Detail:
  This version now builds with cc-5_45. Note: it has not been verified as
  actually functioning correctly.
Admin:
  Tested in DSL Baseline build.

Version 0.31. Tagged as 'NetTime-0_31'

  * Now builds with the latest TCPIPLibs.
Detail:
  * All of the !Mk... files are required to pass TARGET into the makefile.
  * The new TCPIPLibs use the standard types.h from RISC_OSLib if it is
    available. This defines u_int32 types using int rather than long, which
    TCPIPLibs uses. Thus, NetTime needed a little tweaking to adjust for
    this change.
Admin:
  Builds.
  Requires TCPIPLibs-5_42 or later.

Version 0.30. Tagged as 'NetTime-0_30'

Detail:
  A non-existent header file was being included to define the SWI chunk
    information.  This is no longer done.
  Build tidied up a bit.
Admin:
  Built a ROM module and tested fake linking it.
  Built a RAM module (it'll fail due to lack of Messages file though)

Version 0.29. Tagged as 'NetTime-0_29'

Version 0.28. Tagged as 'NetTime-0_28'

Detail:
	Last checkin did not include the file adst.o in the makefile, meaning
that a routine was not being linked against.
	Updated the Functional Specification with the latest changes.
	c.module now uses the global countries header file to get Australia#s
territory number, rather than using the hard-coded value of 29.
Admin:
	Tested.

Version 0.26. Tagged as 'NetTime-0_26'

Detail:
  * If NetTime$DST is set to Australian, instead of European, the
    Australian rules for daylight saving time are applied.
Admin:
  * Tested, and appears to work as expected.
  * Included documentation from Australian legislature detailing the
    changes for DST, and the special case for 2000. Not really required
    for projects at present, but useful to be correct nonetheless.

Version 0.25. Tagged as 'NetTime-0_25'

Detail:
  Protocol auto determination when using a time server specified off
   a smart card was only being done if the time server on the
   smart card was a packed IP address.  If the time server was a
   string, auto determination wasn't being done.  Behaviour changed
   so that auto determination will be done for any smart card format.

Admin:
  Tested on Customer F5 softload.
  Fixes Pace 1618


Version 0.24. Tagged as 'NetTime-0_24'

Now uses "low delay" type of service for outgoing datagrams.

Version 0.23. Tagged as 'NetTime-0_23'

Precision of internal arithmetic and clock reading improved. Now possible to maintain time to within a couple of milliseconds (given frequent enough polling).

Version 0.22. Tagged as 'NetTime-0_22'

If the time server came from NetTime$Server or NCMA, both NTP and Time/UDP
will be tried.
Maximum adjustment rate limited to +/-1%, rather than 10%.

Version 0.21. Tagged as 'NetTime-0_21'

Detail:
  If Inet$NTPServer is set, NetTime will use NTP (unicast SNTP, as per RFC2030).
  Otherwise it will use the UDP time protocol (it used to use TCP).

  NetTime will use the RTCAdjust algorithms to adjust the centisecond clock
  rate by +/-10% to get in step with the time server. By polling every 30
  minutes on a LAN, time will not deviate by more than 0.1 seconds once
  synchronised. It interoperates with RTCAdjust and the hardware RTC to ensure
  correct operation on a machine with RTC.

  NetTime has returned to polling every 30 minutes. This may defeat PPP
  timeouts - a suggested fix is to change PPP to only have TCP traffic keep the
  link up, or to have a separate user inactivity timeout (like Lazarus).

  The system variable NetTime$DST controls DST behaviour. Set to "European"
  (the default) to use European rules. Set it to something like
  "03-May-2000.01:00:00; 14-Sep-2000.01:00:00" to specify this year's start
  and end dates (in UTC). Set to "No" to make it leave DST alone.

  If a DHCP reply arrives while NetTime is loaded, it will check for the time
  offset option - if present, the timezone will be configured.

  New command *NetTime_Status gives information about the clock adjustments.

  NetTime attempts to start operation immediately on a non-dialup machine -
  removes the need for a *NetTime_Kick after softloading. *NetTime_Kick will
  still be required after, say, setting Inet$NTPServer manually.

Admin:
  Tested on desktop systems and STBs. NTP operation on an NC will require a
  way of finding out an NTP server - possibly redefining TIME_SERVER smartcard
  tag to be NTP? Anycast operation may also be worth adding.

Version 0.20. Tagged as 'NetTime-0_20'

NetTime now looks at the Inet$TimeServer system variable before looking at the NetTime$Server variable or the smartcard.

 	This fixes a bug where the server was not getting set if there was
not a time server on the smartcard (or it was present but was not a valid
DNS) as the Inet$TimeServer was not being looked at.

Detail:
 	When attempting to locate a server to gain the time from, it now does
it on the following order:
 	* Checks the Inet$TimeServer system variable first;
 	* If not present, checks the NetTime$Server system variable;
 	* If not present, finally attmepts to get the server off the
smartcard.

Admin:
 	Tested as a softload on an ethernet Customer M box.

Version 0.16. Tagged as 'NetTime-0_16'

Changed to avoid reliance on timestamp but rather to use the defined way of getting a consistent date in the module - use the VersionNum generated string

	* Added StripDepnd support to makefile

	* Date-stamped the CMHG header file to ensure that the new
	  version of the module contains the correct date in the version
	  string

Version 0.15. Tagged as 'NetTime-0_15'

Prevented from doing half-hour time changes, unless PERIODICAL_TIME_CHECKS is defined. If you must keep adjusting the time, it would be a good idea to make sure it changes gradually, like RTCAdjust does it. These checks were preventing PPP from timing out.

Version 0.14. Tagged as 'NetTime-0_14'

Added a header file, debug.h, that is used to redefine some debuglib macros if a non-debug build is being built. This is to allow the module to be built on machines where debuglib is not present.

Validated by building both the old (v0.12) and new versions and ensuring that
the non-debug versions were identical. Also checked that the debug version
still built.

Version 0.13. Tagged as 'NetTime-0_13'

Version 0.12. Tagged as 'NetTime-0_12'

Automatically switch into and out of Daylight Savings for Europe in 1998-2001, if EUROPEAN_DST_BODGE is defined.

Version 0.11. Tagged as 'NetTime-0_11'

Version 0.10. Tagged as 'NetTime-0_10'

  New state (DNSWait) and internal transition (DoDNS) added.
  Machine now support internal transitions, and DoDNS is injected by
  transitions that previously called tryconnect() to start lookup.
  State DNSWait polls resolver every second while it returns EINPROGRESS.
  Resolver interface (dns.c) taken from Acorn FTP fetcher.

Version 0.09. Tagged as 'NetTime-0_09'

Version 0.08. Tagged as 'NetTime-0_08'

Version 0.07. Tagged as 'NetTime-0_07'

Version 0.06. Tagged as 'NetTime-0_06'

Version 0.05. Not tagged