[
https://issues.apache.org/jira/browse/DERBY-3116?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Knut Anders Hatlen updated DERBY-3116:
--------------------------------------
Attachment: d3116-1.diff
I haven't run any tests on it yet, but this patch seems to make totalSpace have
the correct value for all pages created in unit/T_RawStoreFactory.unit
(verified by printing and comparing totalSpace to getMaxFreeSpace()). Will
start a full run of regression tests and report back. I'll also see if I can
add some asserts (since I know of no other way to expose the bug).
What this patch does, is:
- in AllocPage.createPage(), set borrowedSpace before super.createPage() is
called so that getMaxFreeSpace() returns the correct value when totalSpace is
initialized
- in CachedPage.setPageArray(), call usePageBuffer() also when the old buffer
is reused. This ensures that totalSpace is recalculated when a page object is
reused.
> totalSpace not properly initialized in AllocPage
> ------------------------------------------------
>
> Key: DERBY-3116
> URL: https://issues.apache.org/jira/browse/DERBY-3116
> Project: Derby
> Issue Type: Bug
> Components: Store
> Affects Versions: 10.4.0.0
> Reporter: Knut Anders Hatlen
> Priority: Minor
> Attachments: d3116-1.diff
>
>
> There are some problems with the initialization of totalSpace in AllocPage.
> It is initialized in StoredPage.initSpace() which is again called from
> StoredPage.usePageBuffer(), and it is set to the value returned from
> AllocPage.getMaxFreeSpace(). The problems are:
> 1) The calculation in getMaxFreeSpace() uses borrowedSpace, but when
> createIdentity() is called on an AllocPage, borrowedSpace has not been
> initialized when getMaxFreeSpace() is called and the calculated size is wrong.
> 2) When a page object is reused, usePageBuffer() is only called if a new
> byte array must be allocated (because the new page has a different size than
> the old page). This means that the totalSize field gets the same value as in
> the old page if their sizes are equal, which is not necessarily correct.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
