Revert erronous index to a coherent state

[benlabbe]

Could we imagine a process that reverts the NindIndex to its last coherent state ?

Context

for some reason in the documents I try ton index in Nind , I get systematically an exception.

~:$  indexXmlMultimedia corpus_xml/*.mult
...
Indexing corpus_xml/02148_HORZ-ISBL_lot_A_-_EXHIBIT_F_-_Annex_F3_-_rev0.xml.mult (2148/2554, 84%)
terminate called after throwing an instance of 'latecon::nindex::OutWriteBufferException'
  what():  Out write buffer error
Abandon (core dumped)

Since my documents are large (frequently more than 200 pages), the indexation takes nearly 6 hours to reach this error at 84% of the corpus.

First of all, I don't know yet what causes this OutWriteBufferException. The .mult file seems to be erroneous, because when it is excluded, the corpus indexation can complete to its end.

Sadely, this needs a restart of the indexation from the top, because after the OutWriteBufferException, the Nind index files are now corrupted. An indexation on the corrupted Nind files causes a NindPadFileException that doesn't even tell on which file the corruption was found !!

terminate called after throwing an instance of 'latecon::nindex::NindPadFileException'
  what():  Nind Pad error

Propositions

1. How could we ensure that the NindPadFileException tells us on which file the corruption is found ?
   • The class heritage of NindPadFileException from FileException and std::runtime_error does not seems to overwrite the what() method which is supposed to return a context message containing the file on-which the corruption was found.
2. Could we imagine a process that rolls back the last document indexation, so that the Nind files come back to a safe state ?
   • This would allow the restart of the indexing with the next .mult files up to the end of the corpus, instead of restarting from the top.

Calling for help
@kleag ?? @jys ?? could you help here ?