Is there a bootstrap 're-sampling the residuals' equivalent for GLM? In linear regression, I have read of a non-parametric bootstrap being done by 're-sampling the residuals (errors)'. The general idea being that you perturb the mean response by simulated values of the residuals, and take these perturbed values as your re-sample.
I am wondering if there is any equivalent approach for generalized linear models (GLM). 
Since there isn't an error term immediately incorporated into the linear predictor, I'm failing to see how this would be done. The closest equivalent I could think of would be to add some level of Gaussian noise to the linear predictor -- but this seems different enough to go by another name (if such an approach is done at all).
Specifically, I am curious about:


*

*Whether 're-sampling the residuals' done in a GLM setting, and if so, are there other names this approach is commonly called?

*Examples of where this approach has been applied in this setting (e.g. references to journal articles, books, etc.; with the ideal being those materials that discuss the advantages and drawbacks of such an approach).

 A: Yes.
The generalization is most clear if you think of linear regression not with an error term, but as a model for the conditional distribution of $Y \mid X$:
$$ Y \mid X \sim Normal(X \beta, \sigma) $$
When using the parametric bootstrap, we can think of our new $y_i$'s as samples from these conditional distributions, one for each $x_i$. 
This generalizes directly to generalized linear models. For example, logistic regression:
$$ Y \mid X \sim Bernoulli \left( p=\frac{1}{1 + e^{X \beta}} \right) $$
or Poisson regression:
$$ Y \mid X \sim Poisson(\lambda = e^{X \beta}) $$
In each case the parametric bootstrap is the same, we sample a new $y_i$ from the estimated conditional distribution of $Y \mid x_i$.
In the case of linear regression, this is mathematically equivalent to sampling from an error distribution and then adding on the linear predictor, but this error term distribution fails to generalize past the linear regression case (and, I'd argue, this makes the error term description of linear regression somewhat inferior to the conditional distribution description).
