To summarize:

In VFP 7 : oldval(), curval(), eval() screen output is same, some comparison determines that values are different

In VFP 6: oldval(), curval(), eval() screen output is different, comparison determines that their values are same.

I tried your suggestion to exclude corrupted data but havent got consistent results in VFP 7.

At record save time, I need to determine are the table fields changed
by other user to avoid overwriting its changes.

For this, I use code from Doug Hennig error handling paper by scanning all table fields.
However, in VFP 7, if numeric field is not changed at all but corrupted, this code erroneously
reports that data was changed by other user and thus prevents current user to save data its changes.

How to fix this ?

for lnI = 1 to fcount(m.lAlias)
  lcField      = field(lnI)
  llOtherUser  = oldval(lcField)   <> curval(lcField)
  llThisUser   = evaluate(lcField) <> oldval(lcField)
  llSameChange = evaluate(lcField) == curval(lcField)
  do case
    case llOtherUser and !llThisUser
    * Another user edited this field but this user didn't, so 
    * discard current changes. 
    * In VFP 7, it occurs also if numeric field is corrupted. Why ?
.. etc

>Yes, it is interesting.
>Corrupted data is represented differently in VFP6 and in VFP7.
>
>In VFP6 I see something like this (for Windows NT, for Windows98 it is slightly different):
>oldval(lcField) *****.*****
>curval(lcField) |*****.*****
>evaluate(lcField) x0{******.***
>
>So, they are different in VFP6.
>
>
>But in VFP7 they look exactly as in the table:
>oldval(lcField) *
>curval(lcField) *
>evaluate(lcField) *
>
>So, they are the same in VFP7.
>
>*---------------------
>I do not know how to resolve this issue directly on FoxPro level, but for your particular situation I would use some kind of obvious comparison, like:
>
>?evalute(lcField)>1E10
>
>to see whether it is a valid number.
>
>
>
>
>>Yes.
>>
>>But new in VFP 7, the oldval() and curval() for corrupted fields are different.
>>This makes not possible to save a record if it contains a currupted field
>>and remote updating is used.
>>Is it possible to get correct values in llOtherUser and llThisUser ?
>>
>>

>>set multilocks on
>>create table test ( test n(1) )
>>insert into test values (10)
>>use in test
>>use test
>>cursorsetprop( 'buffering', 5 )
>>lcField      = field(1)
>>llOtherUser  = oldval(lcField)   <> curval(lcField)
>>llThisUser   = evaluate(lcField) <> oldval(lcField)
>>* Why llOtherUser and llThisUser are both .T. ?
>>assert !llOtherUSer
>>assert !llThisUSer
>>

>>
>>
>>>The same behaviour you can find in at least in FoxPro2.5 for DOS and in VFP6.
>>>
>>>>I encountered the following two anomalies trying to use VFP 7.
>>>>Can you confirm, are they VFP bugs or not.
>>>>
>>>>Bug 1
>>>>VFP Error does not occur if insert command causes overflow
>>>>

>>>>create table test ( test n(1) )
>>>>insert into test values (10)
>>>>

>>>>
>>>>
>>>>Bug 2
>>>>
>>>>In REPLACE, VFP Error occurs, but data is corrupted
>>>>
>>>>

>>>>create table test ( test n(1) )
>>>>APPEND BLANK
>>>>on error note
>>>>* this line produces error but data is corrupted.
>>>>REPLACE test WITH 10
>>>>on error
>>>>SET ASSERTS ON
>>>>* this assertion fails.
>>>>ASSERT test=0
>>>>