Re: [sqlite] Problems encountered while upgrading Sqlite from 3.6.7 to 3.7.9
I've outputted the results of _sqlite3_libversion and they have verified that the OBJ files are self contained. However, doing so has resulted in one new piece of data. The two versions are: 3.6.17 3.7.9 I don't know if that information makes any difference, but you should have all potentially relevant information. I did add an assertion which runs during initialization to ensure that the version is as expected. I have confirmed that both the shell application and the full application concur on version numbers. On Tue, Jan 10, 2012 at 10:35 AM, John Elrick wrote: > On Tue, Jan 10, 2012 at 10:14 AM, Richard Hipp wrote: >> On Tue, Jan 10, 2012 at 10:08 AM, John Elrick >> wrote: >> >>> On Tue, Jan 10, 2012 at 8:31 AM, Richard Hipp wrote: >>> > On Tue, Jan 10, 2012 at 8:14 AM, John Elrick >> >wrote: >>> > >>> >> I will have to get back to you on this. While running tests against >>> >> isolated queries, I noticed an unusual circumstance. When I isolate >>> >> the queries into a test program, running through our present >>> >> libraries, 3.7.9 is roughly 4 times faster executing the exact same >>> >> queries where it is running roughly 5 times slower in the context of >>> >> the application. As those queries do not execute in the same order, I >>> >> suspect that page swapping and caching issues may be involved. I'm >>> >> proceeding on that assumption. >>> >> >>> > >>> > >>> > Are you using the same compile-time options when building your >>> application >>> > as were used when building the shell program? >>> >>> Yes. >>> >> >> Have you run "SELECT sqlite_source_id();" to verify that your build is >> really picking up the version of SQLite that you think it is? >> >> Do you have code like this in your application: >> >> assert( sqlite3_libversion_number()==SQLITE_VERSION_NUMBER ); >> assert( strcmp(sqlite3_sourceid(),SQLITE_SOURCE_ID)==0 ); >> assert( strcmp(sqlite3_libversion(),SQLITE_VERSION)==0 ); >> >> To verify that your SQLite source code and "sqlite3.h" header file match? > > Nope. I assume that the resulting .OBJ files are self contained. > I'll run the tests you suggest just to make certain. -- John Elrick Fenestra Technologies 540-868-1377 ___ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
Re: [sqlite] Problems encountered while upgrading Sqlite from 3.6.7 to 3.7.9
On Tue, Jan 10, 2012 at 10:14 AM, Richard Hipp wrote: > On Tue, Jan 10, 2012 at 10:08 AM, John Elrick wrote: > >> On Tue, Jan 10, 2012 at 8:31 AM, Richard Hipp wrote: >> > On Tue, Jan 10, 2012 at 8:14 AM, John Elrick > >wrote: >> > >> >> I will have to get back to you on this. While running tests against >> >> isolated queries, I noticed an unusual circumstance. When I isolate >> >> the queries into a test program, running through our present >> >> libraries, 3.7.9 is roughly 4 times faster executing the exact same >> >> queries where it is running roughly 5 times slower in the context of >> >> the application. As those queries do not execute in the same order, I >> >> suspect that page swapping and caching issues may be involved. I'm >> >> proceeding on that assumption. >> >> >> > >> > >> > Are you using the same compile-time options when building your >> application >> > as were used when building the shell program? >> >> Yes. >> > > Have you run "SELECT sqlite_source_id();" to verify that your build is > really picking up the version of SQLite that you think it is? > > Do you have code like this in your application: > > assert( sqlite3_libversion_number()==SQLITE_VERSION_NUMBER ); > assert( strcmp(sqlite3_sourceid(),SQLITE_SOURCE_ID)==0 ); > assert( strcmp(sqlite3_libversion(),SQLITE_VERSION)==0 ); > > To verify that your SQLite source code and "sqlite3.h" header file match? Nope. I assume that the resulting .OBJ files are self contained. I'll run the tests you suggest just to make certain. ___ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
Re: [sqlite] Problems encountered while upgrading Sqlite from 3.6.7 to 3.7.9
On Tue, Jan 10, 2012 at 10:08 AM, John Elrick wrote: > On Tue, Jan 10, 2012 at 8:31 AM, Richard Hipp wrote: > > On Tue, Jan 10, 2012 at 8:14 AM, John Elrick >wrote: > > > >> I will have to get back to you on this. While running tests against > >> isolated queries, I noticed an unusual circumstance. When I isolate > >> the queries into a test program, running through our present > >> libraries, 3.7.9 is roughly 4 times faster executing the exact same > >> queries where it is running roughly 5 times slower in the context of > >> the application. As those queries do not execute in the same order, I > >> suspect that page swapping and caching issues may be involved. I'm > >> proceeding on that assumption. > >> > > > > > > Are you using the same compile-time options when building your > application > > as were used when building the shell program? > > Yes. > Have you run "SELECT sqlite_source_id();" to verify that your build is really picking up the version of SQLite that you think it is? Do you have code like this in your application: assert( sqlite3_libversion_number()==SQLITE_VERSION_NUMBER ); assert( strcmp(sqlite3_sourceid(),SQLITE_SOURCE_ID)==0 ); assert( strcmp(sqlite3_libversion(),SQLITE_VERSION)==0 ); To verify that your SQLite source code and "sqlite3.h" header file match? > ___ > sqlite-users mailing list > sqlite-users@sqlite.org > http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users > -- D. Richard Hipp d...@sqlite.org ___ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
Re: [sqlite] Problems encountered while upgrading Sqlite from 3.6.7 to 3.7.9
On Tue, Jan 10, 2012 at 8:31 AM, Richard Hipp wrote: > On Tue, Jan 10, 2012 at 8:14 AM, John Elrick wrote: > >> I will have to get back to you on this. While running tests against >> isolated queries, I noticed an unusual circumstance. When I isolate >> the queries into a test program, running through our present >> libraries, 3.7.9 is roughly 4 times faster executing the exact same >> queries where it is running roughly 5 times slower in the context of >> the application. As those queries do not execute in the same order, I >> suspect that page swapping and caching issues may be involved. I'm >> proceeding on that assumption. >> > > > Are you using the same compile-time options when building your application > as were used when building the shell program? Yes. ___ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
Re: [sqlite] Problems encountered while upgrading Sqlite from 3.6.7 to 3.7.9
On Tue, Jan 10, 2012 at 8:14 AM, John Elrick wrote: > I will have to get back to you on this. While running tests against > isolated queries, I noticed an unusual circumstance. When I isolate > the queries into a test program, running through our present > libraries, 3.7.9 is roughly 4 times faster executing the exact same > queries where it is running roughly 5 times slower in the context of > the application. As those queries do not execute in the same order, I > suspect that page swapping and caching issues may be involved. I'm > proceeding on that assumption. > Are you using the same compile-time options when building your application as were used when building the shell program? -- D. Richard Hipp d...@sqlite.org ___ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
Re: [sqlite] Problems encountered while upgrading Sqlite from 3.6.7 to 3.7.9
I will have to get back to you on this. While running tests against isolated queries, I noticed an unusual circumstance. When I isolate the queries into a test program, running through our present libraries, 3.7.9 is roughly 4 times faster executing the exact same queries where it is running roughly 5 times slower in the context of the application. As those queries do not execute in the same order, I suspect that page swapping and caching issues may be involved. I'm proceeding on that assumption. On Mon, Jan 9, 2012 at 12:11 PM, Richard Hipp wrote: > Thank you for sending the sample database file. > > Rather than forcing me to understand your log database, and fill in > parameters with reasonable values, I wonder if you could simply email one > or more queries (with the parameters filled in) which you find to be slower > in 3.7.9 versus 3.6.7? > > On Mon, Jan 9, 2012 at 11:13 AM, Richard Hipp wrote: > >> Please send private email to d...@sqlite.org. CC to d...@sqlite.org and >> j...@sqlite.org. >> >> >> On Mon, Jan 9, 2012 at 11:09 AM, John Elrick wrote: >> >>> Sorry "The combined files are 2.4 MB zipped" >>> >>> On Mon, Jan 9, 2012 at 11:09 AM, John Elrick >>> wrote: >>> > Richard, >>> > >>> > I have the information available for you. Unfortunately, it appears >>> > that I cannot actually reduce the case, however, the issues appear to >>> > be fairly confined. >>> > >>> > The test database is 6 MB and I have a database showing the >>> > performance tracking for each of the queries which is about 140KB. >>> > The combined files are 2.4 MB. How would you prefer I send this file >>> > to you? >>> > >>> > On Fri, Jan 6, 2012 at 4:51 PM, John Elrick >>> wrote: >>> >> Yes. Look for it on Monday. I'm about to leave for the day. The >>> >> queries are automatically generated and will require some mining to >>> >> get specific examples. I should be able to reduce the range to a >>> >> simple repeatable case but it may take a while. >>> >> >>> >> Thanks very much and have a great weekend. >>> >> >>> >> On Fri, Jan 6, 2012 at 4:42 PM, Richard Hipp wrote: >>> >>> Can you publish for us: >>> >>> >>> >>> (1) A sample database >>> >>> (2) The specific queries that are causing your problems >>> >>> >>> >>> So that we can investigate further? >>> >>> >>> >>> On Fri, Jan 6, 2012 at 4:31 PM, John Elrick >> >wrote: >>> >>> >>> Background >>> = >>> We have been using Sqlite3 in a project since 2007. Because of the >>> release cycles involved, we normally upgrade infrequently. We >>> recently upgraded from 3.6.7 to 3.7.9 and encountered two serious >>> issues. We need assistance in narrowing down the causes. >>> >>> We are developing using Delphi. We are using the amalgamation and >>> are >>> compiling it using the free Borland BCC complier. The result is an >>> obj file which is statically linked into our application. We have >>> been using this technique since we began using Sqlite. >>> >>> In the tests below, the strategy for changing between versions is to >>> change which obj file is statically linked into the application. The >>> application was recompiled between each test. >>> >>> Issues Encountered >>> === >>> >>> 1. After changing there is an order of magnitude performance >>> reduction. Using GP Profile to monitor operations, here are the >>> results for identical application uses: >>> >>> Operation executes _sqlite3_step 49,152 times >>> >>> 3.6.7 time 5.24 seconds >>> 3.7.9 time 41.19 seconds >>> >>> 2. During testing we encountered a bug which had not been >>> encountered >>> before. Research by our CTO established that in 3.7.9 a row was >>> returned from this query which contained empty values. In 3.6.7 >>> there >>> were no results from the query. >>> >>> Summation >>> = >>> Given how throughly Sqlite is tested I have difficulty believing >>> there >>> is an actual bug in Sqlite. I believe an interaction is occurring >>> which is a result of changes to the Sqlite code base which are >>> producing unexpected resulted when compiled into our application. To >>> help resolve the problem, I would like to know if there is an >>> historical record of the amalgamations. If so, I can carry out >>> testing to determine in what version the problem manifests. >>> >>> If anyone has any other recommendations I would be pleased to hear >>> them. For the moment we are reverting to 3.6.7 as a temporary >>> solution. >>> >>> -- >>> John Elrick >>> Fenestra Technologies >>> 540-868-1377 >>> ___ >>> sqlite-users mailing list >>> sqlite-users@sqlite.org >>> http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users >>> >>> >>> >>> >>> >>> >>> >>> >>> -
Re: [sqlite] Problems encountered while upgrading Sqlite from 3.6.7 to 3.7.9
Thank you for sending the sample database file. Rather than forcing me to understand your log database, and fill in parameters with reasonable values, I wonder if you could simply email one or more queries (with the parameters filled in) which you find to be slower in 3.7.9 versus 3.6.7? On Mon, Jan 9, 2012 at 11:13 AM, Richard Hipp wrote: > Please send private email to d...@sqlite.org. CC to d...@sqlite.org and > j...@sqlite.org. > > > On Mon, Jan 9, 2012 at 11:09 AM, John Elrick wrote: > >> Sorry "The combined files are 2.4 MB zipped" >> >> On Mon, Jan 9, 2012 at 11:09 AM, John Elrick >> wrote: >> > Richard, >> > >> > I have the information available for you. Unfortunately, it appears >> > that I cannot actually reduce the case, however, the issues appear to >> > be fairly confined. >> > >> > The test database is 6 MB and I have a database showing the >> > performance tracking for each of the queries which is about 140KB. >> > The combined files are 2.4 MB. How would you prefer I send this file >> > to you? >> > >> > On Fri, Jan 6, 2012 at 4:51 PM, John Elrick >> wrote: >> >> Yes. Look for it on Monday. I'm about to leave for the day. The >> >> queries are automatically generated and will require some mining to >> >> get specific examples. I should be able to reduce the range to a >> >> simple repeatable case but it may take a while. >> >> >> >> Thanks very much and have a great weekend. >> >> >> >> On Fri, Jan 6, 2012 at 4:42 PM, Richard Hipp wrote: >> >>> Can you publish for us: >> >>> >> >>> (1) A sample database >> >>> (2) The specific queries that are causing your problems >> >>> >> >>> So that we can investigate further? >> >>> >> >>> On Fri, Jan 6, 2012 at 4:31 PM, John Elrick > >wrote: >> >>> >> Background >> = >> We have been using Sqlite3 in a project since 2007. Because of the >> release cycles involved, we normally upgrade infrequently. We >> recently upgraded from 3.6.7 to 3.7.9 and encountered two serious >> issues. We need assistance in narrowing down the causes. >> >> We are developing using Delphi. We are using the amalgamation and >> are >> compiling it using the free Borland BCC complier. The result is an >> obj file which is statically linked into our application. We have >> been using this technique since we began using Sqlite. >> >> In the tests below, the strategy for changing between versions is to >> change which obj file is statically linked into the application. The >> application was recompiled between each test. >> >> Issues Encountered >> === >> >> 1. After changing there is an order of magnitude performance >> reduction. Using GP Profile to monitor operations, here are the >> results for identical application uses: >> >> Operation executes _sqlite3_step 49,152 times >> >> 3.6.7 time 5.24 seconds >> 3.7.9 time 41.19 seconds >> >> 2. During testing we encountered a bug which had not been >> encountered >> before. Research by our CTO established that in 3.7.9 a row was >> returned from this query which contained empty values. In 3.6.7 >> there >> were no results from the query. >> >> Summation >> = >> Given how throughly Sqlite is tested I have difficulty believing >> there >> is an actual bug in Sqlite. I believe an interaction is occurring >> which is a result of changes to the Sqlite code base which are >> producing unexpected resulted when compiled into our application. To >> help resolve the problem, I would like to know if there is an >> historical record of the amalgamations. If so, I can carry out >> testing to determine in what version the problem manifests. >> >> If anyone has any other recommendations I would be pleased to hear >> them. For the moment we are reverting to 3.6.7 as a temporary >> solution. >> >> -- >> John Elrick >> Fenestra Technologies >> 540-868-1377 >> ___ >> sqlite-users mailing list >> sqlite-users@sqlite.org >> http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users >> >> >>> >> >>> >> >>> >> >>> -- >> >>> D. Richard Hipp >> >>> d...@sqlite.org >> >>> ___ >> >>> sqlite-users mailing list >> >>> sqlite-users@sqlite.org >> >>> http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users >> >> >> >> >> >> >> >> -- >> >> John Elrick >> >> Fenestra Technologies >> >> 540-868-1377 >> > >> > >> > >> > -- >> > John Elrick >> > Fenestra Technologies >> > 540-868-1377 >> >> >> >> -- >> John Elrick >> Fenestra Technologies >> 540-868-1377 >> ___ >> sqlite-users mailing list >> sqlite-users@sqlite.org >> http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users >> > > > > -- > D. Richard
Re: [sqlite] Problems encountered while upgrading Sqlite from 3.6.7 to 3.7.9
Please send private email to d...@sqlite.org. CC to d...@sqlite.org and j...@sqlite.org. On Mon, Jan 9, 2012 at 11:09 AM, John Elrick wrote: > Sorry "The combined files are 2.4 MB zipped" > > On Mon, Jan 9, 2012 at 11:09 AM, John Elrick > wrote: > > Richard, > > > > I have the information available for you. Unfortunately, it appears > > that I cannot actually reduce the case, however, the issues appear to > > be fairly confined. > > > > The test database is 6 MB and I have a database showing the > > performance tracking for each of the queries which is about 140KB. > > The combined files are 2.4 MB. How would you prefer I send this file > > to you? > > > > On Fri, Jan 6, 2012 at 4:51 PM, John Elrick > wrote: > >> Yes. Look for it on Monday. I'm about to leave for the day. The > >> queries are automatically generated and will require some mining to > >> get specific examples. I should be able to reduce the range to a > >> simple repeatable case but it may take a while. > >> > >> Thanks very much and have a great weekend. > >> > >> On Fri, Jan 6, 2012 at 4:42 PM, Richard Hipp wrote: > >>> Can you publish for us: > >>> > >>> (1) A sample database > >>> (2) The specific queries that are causing your problems > >>> > >>> So that we can investigate further? > >>> > >>> On Fri, Jan 6, 2012 at 4:31 PM, John Elrick >wrote: > >>> > Background > = > We have been using Sqlite3 in a project since 2007. Because of the > release cycles involved, we normally upgrade infrequently. We > recently upgraded from 3.6.7 to 3.7.9 and encountered two serious > issues. We need assistance in narrowing down the causes. > > We are developing using Delphi. We are using the amalgamation and are > compiling it using the free Borland BCC complier. The result is an > obj file which is statically linked into our application. We have > been using this technique since we began using Sqlite. > > In the tests below, the strategy for changing between versions is to > change which obj file is statically linked into the application. The > application was recompiled between each test. > > Issues Encountered > === > > 1. After changing there is an order of magnitude performance > reduction. Using GP Profile to monitor operations, here are the > results for identical application uses: > > Operation executes _sqlite3_step 49,152 times > > 3.6.7 time 5.24 seconds > 3.7.9 time 41.19 seconds > > 2. During testing we encountered a bug which had not been encountered > before. Research by our CTO established that in 3.7.9 a row was > returned from this query which contained empty values. In 3.6.7 there > were no results from the query. > > Summation > = > Given how throughly Sqlite is tested I have difficulty believing there > is an actual bug in Sqlite. I believe an interaction is occurring > which is a result of changes to the Sqlite code base which are > producing unexpected resulted when compiled into our application. To > help resolve the problem, I would like to know if there is an > historical record of the amalgamations. If so, I can carry out > testing to determine in what version the problem manifests. > > If anyone has any other recommendations I would be pleased to hear > them. For the moment we are reverting to 3.6.7 as a temporary > solution. > > -- > John Elrick > Fenestra Technologies > 540-868-1377 > ___ > sqlite-users mailing list > sqlite-users@sqlite.org > http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users > > >>> > >>> > >>> > >>> -- > >>> D. Richard Hipp > >>> d...@sqlite.org > >>> ___ > >>> sqlite-users mailing list > >>> sqlite-users@sqlite.org > >>> http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users > >> > >> > >> > >> -- > >> John Elrick > >> Fenestra Technologies > >> 540-868-1377 > > > > > > > > -- > > John Elrick > > Fenestra Technologies > > 540-868-1377 > > > > -- > John Elrick > Fenestra Technologies > 540-868-1377 > ___ > sqlite-users mailing list > sqlite-users@sqlite.org > http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users > -- D. Richard Hipp d...@sqlite.org ___ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
Re: [sqlite] Problems encountered while upgrading Sqlite from 3.6.7 to 3.7.9
Sorry "The combined files are 2.4 MB zipped" On Mon, Jan 9, 2012 at 11:09 AM, John Elrick wrote: > Richard, > > I have the information available for you. Unfortunately, it appears > that I cannot actually reduce the case, however, the issues appear to > be fairly confined. > > The test database is 6 MB and I have a database showing the > performance tracking for each of the queries which is about 140KB. > The combined files are 2.4 MB. How would you prefer I send this file > to you? > > On Fri, Jan 6, 2012 at 4:51 PM, John Elrick wrote: >> Yes. Look for it on Monday. I'm about to leave for the day. The >> queries are automatically generated and will require some mining to >> get specific examples. I should be able to reduce the range to a >> simple repeatable case but it may take a while. >> >> Thanks very much and have a great weekend. >> >> On Fri, Jan 6, 2012 at 4:42 PM, Richard Hipp wrote: >>> Can you publish for us: >>> >>> (1) A sample database >>> (2) The specific queries that are causing your problems >>> >>> So that we can investigate further? >>> >>> On Fri, Jan 6, 2012 at 4:31 PM, John Elrick wrote: >>> Background = We have been using Sqlite3 in a project since 2007. Because of the release cycles involved, we normally upgrade infrequently. We recently upgraded from 3.6.7 to 3.7.9 and encountered two serious issues. We need assistance in narrowing down the causes. We are developing using Delphi. We are using the amalgamation and are compiling it using the free Borland BCC complier. The result is an obj file which is statically linked into our application. We have been using this technique since we began using Sqlite. In the tests below, the strategy for changing between versions is to change which obj file is statically linked into the application. The application was recompiled between each test. Issues Encountered === 1. After changing there is an order of magnitude performance reduction. Using GP Profile to monitor operations, here are the results for identical application uses: Operation executes _sqlite3_step 49,152 times 3.6.7 time 5.24 seconds 3.7.9 time 41.19 seconds 2. During testing we encountered a bug which had not been encountered before. Research by our CTO established that in 3.7.9 a row was returned from this query which contained empty values. In 3.6.7 there were no results from the query. Summation = Given how throughly Sqlite is tested I have difficulty believing there is an actual bug in Sqlite. I believe an interaction is occurring which is a result of changes to the Sqlite code base which are producing unexpected resulted when compiled into our application. To help resolve the problem, I would like to know if there is an historical record of the amalgamations. If so, I can carry out testing to determine in what version the problem manifests. If anyone has any other recommendations I would be pleased to hear them. For the moment we are reverting to 3.6.7 as a temporary solution. -- John Elrick Fenestra Technologies 540-868-1377 ___ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users >>> >>> >>> >>> -- >>> D. Richard Hipp >>> d...@sqlite.org >>> ___ >>> sqlite-users mailing list >>> sqlite-users@sqlite.org >>> http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users >> >> >> >> -- >> John Elrick >> Fenestra Technologies >> 540-868-1377 > > > > -- > John Elrick > Fenestra Technologies > 540-868-1377 -- John Elrick Fenestra Technologies 540-868-1377 ___ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
Re: [sqlite] Problems encountered while upgrading Sqlite from 3.6.7 to 3.7.9
Richard, I have the information available for you. Unfortunately, it appears that I cannot actually reduce the case, however, the issues appear to be fairly confined. The test database is 6 MB and I have a database showing the performance tracking for each of the queries which is about 140KB. The combined files are 2.4 MB. How would you prefer I send this file to you? On Fri, Jan 6, 2012 at 4:51 PM, John Elrick wrote: > Yes. Look for it on Monday. I'm about to leave for the day. The > queries are automatically generated and will require some mining to > get specific examples. I should be able to reduce the range to a > simple repeatable case but it may take a while. > > Thanks very much and have a great weekend. > > On Fri, Jan 6, 2012 at 4:42 PM, Richard Hipp wrote: >> Can you publish for us: >> >> (1) A sample database >> (2) The specific queries that are causing your problems >> >> So that we can investigate further? >> >> On Fri, Jan 6, 2012 at 4:31 PM, John Elrick wrote: >> >>> Background >>> = >>> We have been using Sqlite3 in a project since 2007. Because of the >>> release cycles involved, we normally upgrade infrequently. We >>> recently upgraded from 3.6.7 to 3.7.9 and encountered two serious >>> issues. We need assistance in narrowing down the causes. >>> >>> We are developing using Delphi. We are using the amalgamation and are >>> compiling it using the free Borland BCC complier. The result is an >>> obj file which is statically linked into our application. We have >>> been using this technique since we began using Sqlite. >>> >>> In the tests below, the strategy for changing between versions is to >>> change which obj file is statically linked into the application. The >>> application was recompiled between each test. >>> >>> Issues Encountered >>> === >>> >>> 1. After changing there is an order of magnitude performance >>> reduction. Using GP Profile to monitor operations, here are the >>> results for identical application uses: >>> >>> Operation executes _sqlite3_step 49,152 times >>> >>> 3.6.7 time 5.24 seconds >>> 3.7.9 time 41.19 seconds >>> >>> 2. During testing we encountered a bug which had not been encountered >>> before. Research by our CTO established that in 3.7.9 a row was >>> returned from this query which contained empty values. In 3.6.7 there >>> were no results from the query. >>> >>> Summation >>> = >>> Given how throughly Sqlite is tested I have difficulty believing there >>> is an actual bug in Sqlite. I believe an interaction is occurring >>> which is a result of changes to the Sqlite code base which are >>> producing unexpected resulted when compiled into our application. To >>> help resolve the problem, I would like to know if there is an >>> historical record of the amalgamations. If so, I can carry out >>> testing to determine in what version the problem manifests. >>> >>> If anyone has any other recommendations I would be pleased to hear >>> them. For the moment we are reverting to 3.6.7 as a temporary >>> solution. >>> >>> -- >>> John Elrick >>> Fenestra Technologies >>> 540-868-1377 >>> ___ >>> sqlite-users mailing list >>> sqlite-users@sqlite.org >>> http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users >>> >> >> >> >> -- >> D. Richard Hipp >> d...@sqlite.org >> ___ >> sqlite-users mailing list >> sqlite-users@sqlite.org >> http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users > > > > -- > John Elrick > Fenestra Technologies > 540-868-1377 -- John Elrick Fenestra Technologies 540-868-1377 ___ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
Re: [sqlite] Problems encountered while upgrading Sqlite from 3.6.7 to 3.7.9
Yes. Look for it on Monday. I'm about to leave for the day. The queries are automatically generated and will require some mining to get specific examples. I should be able to reduce the range to a simple repeatable case but it may take a while. Thanks very much and have a great weekend. On Fri, Jan 6, 2012 at 4:42 PM, Richard Hipp wrote: > Can you publish for us: > > (1) A sample database > (2) The specific queries that are causing your problems > > So that we can investigate further? > > On Fri, Jan 6, 2012 at 4:31 PM, John Elrick wrote: > >> Background >> = >> We have been using Sqlite3 in a project since 2007. Because of the >> release cycles involved, we normally upgrade infrequently. We >> recently upgraded from 3.6.7 to 3.7.9 and encountered two serious >> issues. We need assistance in narrowing down the causes. >> >> We are developing using Delphi. We are using the amalgamation and are >> compiling it using the free Borland BCC complier. The result is an >> obj file which is statically linked into our application. We have >> been using this technique since we began using Sqlite. >> >> In the tests below, the strategy for changing between versions is to >> change which obj file is statically linked into the application. The >> application was recompiled between each test. >> >> Issues Encountered >> === >> >> 1. After changing there is an order of magnitude performance >> reduction. Using GP Profile to monitor operations, here are the >> results for identical application uses: >> >> Operation executes _sqlite3_step 49,152 times >> >> 3.6.7 time 5.24 seconds >> 3.7.9 time 41.19 seconds >> >> 2. During testing we encountered a bug which had not been encountered >> before. Research by our CTO established that in 3.7.9 a row was >> returned from this query which contained empty values. In 3.6.7 there >> were no results from the query. >> >> Summation >> = >> Given how throughly Sqlite is tested I have difficulty believing there >> is an actual bug in Sqlite. I believe an interaction is occurring >> which is a result of changes to the Sqlite code base which are >> producing unexpected resulted when compiled into our application. To >> help resolve the problem, I would like to know if there is an >> historical record of the amalgamations. If so, I can carry out >> testing to determine in what version the problem manifests. >> >> If anyone has any other recommendations I would be pleased to hear >> them. For the moment we are reverting to 3.6.7 as a temporary >> solution. >> >> -- >> John Elrick >> Fenestra Technologies >> 540-868-1377 >> ___ >> sqlite-users mailing list >> sqlite-users@sqlite.org >> http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users >> > > > > -- > D. Richard Hipp > d...@sqlite.org > ___ > sqlite-users mailing list > sqlite-users@sqlite.org > http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users -- John Elrick Fenestra Technologies 540-868-1377 ___ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
Re: [sqlite] Problems encountered while upgrading Sqlite from 3.6.7 to 3.7.9
Can you publish for us: (1) A sample database (2) The specific queries that are causing your problems So that we can investigate further? On Fri, Jan 6, 2012 at 4:31 PM, John Elrick wrote: > Background > = > We have been using Sqlite3 in a project since 2007. Because of the > release cycles involved, we normally upgrade infrequently. We > recently upgraded from 3.6.7 to 3.7.9 and encountered two serious > issues. We need assistance in narrowing down the causes. > > We are developing using Delphi. We are using the amalgamation and are > compiling it using the free Borland BCC complier. The result is an > obj file which is statically linked into our application. We have > been using this technique since we began using Sqlite. > > In the tests below, the strategy for changing between versions is to > change which obj file is statically linked into the application. The > application was recompiled between each test. > > Issues Encountered > === > > 1. After changing there is an order of magnitude performance > reduction. Using GP Profile to monitor operations, here are the > results for identical application uses: > > Operation executes _sqlite3_step 49,152 times > > 3.6.7 time 5.24 seconds > 3.7.9 time 41.19 seconds > > 2. During testing we encountered a bug which had not been encountered > before. Research by our CTO established that in 3.7.9 a row was > returned from this query which contained empty values. In 3.6.7 there > were no results from the query. > > Summation > = > Given how throughly Sqlite is tested I have difficulty believing there > is an actual bug in Sqlite. I believe an interaction is occurring > which is a result of changes to the Sqlite code base which are > producing unexpected resulted when compiled into our application. To > help resolve the problem, I would like to know if there is an > historical record of the amalgamations. If so, I can carry out > testing to determine in what version the problem manifests. > > If anyone has any other recommendations I would be pleased to hear > them. For the moment we are reverting to 3.6.7 as a temporary > solution. > > -- > John Elrick > Fenestra Technologies > 540-868-1377 > ___ > sqlite-users mailing list > sqlite-users@sqlite.org > http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users > -- D. Richard Hipp d...@sqlite.org ___ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
[sqlite] Problems encountered while upgrading Sqlite from 3.6.7 to 3.7.9
Background = We have been using Sqlite3 in a project since 2007. Because of the release cycles involved, we normally upgrade infrequently. We recently upgraded from 3.6.7 to 3.7.9 and encountered two serious issues. We need assistance in narrowing down the causes. We are developing using Delphi. We are using the amalgamation and are compiling it using the free Borland BCC complier. The result is an obj file which is statically linked into our application. We have been using this technique since we began using Sqlite. In the tests below, the strategy for changing between versions is to change which obj file is statically linked into the application. The application was recompiled between each test. Issues Encountered === 1. After changing there is an order of magnitude performance reduction. Using GP Profile to monitor operations, here are the results for identical application uses: Operation executes _sqlite3_step 49,152 times 3.6.7 time 5.24 seconds 3.7.9 time 41.19 seconds 2. During testing we encountered a bug which had not been encountered before. Research by our CTO established that in 3.7.9 a row was returned from this query which contained empty values. In 3.6.7 there were no results from the query. Summation = Given how throughly Sqlite is tested I have difficulty believing there is an actual bug in Sqlite. I believe an interaction is occurring which is a result of changes to the Sqlite code base which are producing unexpected resulted when compiled into our application. To help resolve the problem, I would like to know if there is an historical record of the amalgamations. If so, I can carry out testing to determine in what version the problem manifests. If anyone has any other recommendations I would be pleased to hear them. For the moment we are reverting to 3.6.7 as a temporary solution. -- John Elrick Fenestra Technologies 540-868-1377 ___ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users